Training-free camera control for pretrained flow-matching video generators is a partial-observation inverse problem: a depth-warped guidance video supplies noisy evidence on a subset of latent sites, which the sampler must reconcile with the pretrained prior. Existing methods struggle to balance the trade-off between trajectory adherence and visual quality and the heuristic guidance-strength tuning lacks robustness. We propose $h$-control, which resolves this dilemma through a structural change to the sampler: each outer hard-replacement guidance step is augmented with an inner-loop block-conditional pseudo-Gibbs refinement on the unobserved complement at the same noise level, with provable convergence to the partial-observation conditional data law. To accelerate convergence on high-dimensional video latents, we exploit their conditional locality, partitioning the unobserved complement into 3D patches, each tracked by a custom mixing indicator that adaptively freezes converged patches. On RealEstate10K and DAVIS, $h$-control attains the best FVD against all seven training-free and training-based competitors, outperforming every training-free baseline on every reported metric.

Systematicity is the capacity to generalize compositionally. Existing benchmarks primarily test behavioral systematicity, while representation-level analyses often leave open whether recovered signals are merely decodable or causally used during generation. We therefore introduce $\mathrm{Bounded\mbox{-}RS}$, an evaluation artifact and protocol for adjudicating these possibilities under bounded hypotheses: unsupported, present-only, causal-use, or underdetermined. The artifact is a controlled semantic-parsing setting with matched recombination tests, exact intervention pairs, and evidence routines for recoverability, interchange, necessity, and repair during generation. We apply $\mathrm{Bounded\mbox{-}RS}$ to the two prominent representation-level explanations of systematic generalization, binding and geometry. Within the family instantiations evaluated here, and under the same behavioral anchor and state semantics, the binding family satisfies the criteria for $\mathit{Causal\mbox{-}use}$, whereas the geometry family is adjudicated as $\mathit{Unsupported}$ under its frozen closure basis. Detector-truthfulness tests do not validate a geometry detector, so the geometry conclusion remains a claim about the tested family rather than about all possible geometric mechanisms. $\mathrm{Bounded\mbox{-}RS}$ thus provides a reusable way to make disciplined representational claims from behavioral systematicity tests.

Recent feed-forward geometry foundation models have demonstrated impressive generalization by recovering depth and poses in a single forward pass. However, these models are typically constrained by a global coordinate frame assumption. This dependency becomes a significant bottleneck for long-context and streaming reconstruction, as it forces the network to maintain an arbitrary temporal origin and handle translation magnitudes that grow unbounded over time. Our solution, which we call $R^3$, employs relative regression. We employ a lightweight MLP to predict confidence-weighted relative constraints. These confidences serve as a unified anchor: weighting losses during training and guiding pose aggregation during inference. $R^3$ supports both full-context offline reconstruction and causal, bounded-memory streaming. Our evaluation in both offline and streaming settings validates the effectiveness of our relative mechanism.

This paper presents an End-to-End learnable radar semantic segmentation network using I/Q echoes as input for Doppler radars, named as $\textbf{EchMo} (\text{\textbf{ech}o-in-\textbf{m}ask-\textbf{o}ut})$. EchMo reinterprets the steps of conventional radar signal processing (RSP) in a deeply learnable manner, including pulse compression (traditionally implemented via matched filtering), moving target indication (MTI), coherent integration (CI), and target detection. Ultimately, it achieves a single compact learnable deep model that goes from I/Q echoes to target semantic segmentation in one step, delivering a streamlined differentiable computational graph and substantial efficiency gains. Experimental results show that, compared to conventional pipeline or deep models based on intermediate radio-frequency (RF) representation, EchMo achieves better results and offers a more interpretable foundation. EchMo is the first pulse-Doppler I/Q echo-based end-to-end radar semantic segmentation deep architecture, which is a milestone for the fields of RSP, remote sensing, and modern deep learning. The code link for review: https://anonymous.4open.science/anonymize/EchMo-6877.

While text-to-image (T2I) diffusion models exhibit strong semantic disentanglement at the object level, they struggle to localize fine-grained object parts. Although the latent visual features in these models are sufficiently fine-grained, the text–image interaction (cross-attention) fails to effectively exploit this information. This results in coarse and ambiguous localization, particularly for spatially distinct components such as left and right limbs. To address this limitation, we introduce PartConcepts, a mechanism that encodes textual part descriptions into compact, learnable representations. These PartConcept tokens are trained to selectively attend to their corresponding part regions in the image. To evaluate the part-level understanding of PartConcept tokens, we first probe its effectiveness for part segmentation. We then evaluate whether the improved localization enables fine-grained instruction following in T2I generation. On the standard Open-Vocabulary Part Segmentation (OVPS) benchmark, our method outperforms dedicated segmentation baselines, establishing a new state-of-the-art. We further evaluate our method on the more complex Pascal-Part benchmark for part instance segmentation, which contains fine-grained spatial annotations (e.g. left lower arm). Here, we outperform all baselines by a very significant margin. Finally, our qualitative and quantitative results show that this strong localization of PartConcept tokens directly enables fine-grained instruction following in T2I generation: the textual attributes (e.g., color) inherently bind well to corresponding PartConcept tokens, without explicit attribute control mechanisms. This indicates that PartConcept tokens compose seamlessly with other text tokens, highlighting their applicability for fine-grained text-based generative control.

Many machine learning tasks are invariant under the action of a group $G$ of transformations: signal classification can be invariant under translations, image classification under 2D rotations, and spherical-image classification under 3D rotations. The $G$-bispectrum is a principled complete invariant of a signal )retaining all signal's information up to the group action) with proven benefits in machine learning and as a pooling layer in deep networks. However, its deployment has been hampered by high computational cost and a patchwork of group-specific implementations. We present bispectrum, an open-source, fully unit-tested PyTorch library that implements selective $G$-bispectra for seven different group actions, as differentiable modules that can be directly incorporated into machine learning pipelines and deep learning architectures. For finite groups $G$, selectivity reduces the computational cost from $O(|G|^2)$ to $O(|G|)$. For planar rotations, we leverage the disk bispectrum. For spherical 3D rotations, we introduce an augmented selective bispectrum at band-limit $L$ which reduces the cost from $O(L^3)$ to $\Theta(L^2)$ coefficients. We profile the entire library (for

Three-dimensional molecular structure generation is typically performed at the level of individual atoms, yet molecular graph generation techniques often consider *fragments* as their structural units. Building on the advances in *frame-based* protein structure generation, we extend these fragmentation ideas to 3D, treating general molecules as sets of *rigid-body motifs*. Utilising this representation, we employ $\mathrm{SE}(3)$-equivariant generative modelling for *de novo* 3D molecule generation from rigid motifs. In our evaluations, we observe comparable or superior results to state-of-the-art across benchmarks, surpassing it in atom stability on GEOM-Drugs, while yielding a $2\times$ to $10\times$ reduction in generation steps and offering $3.5\times$ compression in molecular representations compared to the standard atom-based methods.

Graph representation learning often relies on message passing or spectral/positional encodings to summarize graph structure, but these indirect summaries can collapse structurally distinct graphs, including non-isomorphic cospectral pairs. We propose A2I, a structural encoding framework that renders BFS--degree-ordered local adjacency patterns as fixed-resolution images and embeds them with a frozen pre-trained vision encoder. The resulting structural tokens are mapped to learnable prototypes and aggregated with node embeddings via a lightweight Transformer. We provide a conditional analysis with two related findings: a stability result, showing that two orderings of the same graph yield embeddings that converge under bounded-degree assumptions, and a separability result, showing that two distinct graphs yield embeddings whose gap grows at least linearly with the difference in their BFS profiles. Empirically, A2I is competitive with recent GNN, Graph Transformer, and structural-encoding baselines on six node-classification benchmarks, with pronounced gains for GCN-based models on heterophilous graphs and under partial feature masking.

Modern neural-network optimizers occupy a three-way design space: online adaptivity, architecture-aware geometry, and acceleration. We introduce AcceleGrad#, an accelerated linear-coupling template that couples an adaptive anchor with a norm-induced sharp, mirror, or linear-minimization geometry branch. The scalar-anchor specialization recovers the accelerated $O(1/K^2)$ rate for smooth convex objectives, while the diagonal-AdaGrad clipped-LMO training variant admits local $\mu$-Kurdyka-Lojasiewicz certificates: a generic conservative $O(K^{-1/4})$ rate and, in a SCION cross-entropy regime, a $O(K^{-1/3})$ last-iterate rate to a self-bounded mini-batch noise level. On image classification and language modeling, the training variant improves over Euclidean AcceleGrad and remains competitive with recent non-Euclidean optimizers. Using stochastic gradients directly in spectral oracles also reduces polar-computation time and optimizer overhead in practice.

Text-based image editing has recently been reinterpreted in large multimodal transformers as conditional generation, where source image tokens are concatenated with text and noise tokens. While effective, this design incurs substantial computational overhead in attention layers. To mitigate this inefficiency, we propose {\em TokenDrop}, an efficient editing framework that shift the computational burden from dense source-token conditioning to a lightweight regularized sampling update. By transferring the influence of source tokens into a closed-form sampling update, the method preserves source consistency with negligible cost. To support this regularized dynamics, we introduce a deviation-guided adaptive masking strategy that selectively drops redundant tokens while maintaining editing fidelity. Across FluxKontext and Qwen-Image-Edit, our training-free method achieves an average 22.4\% improvement in inference speed on PIEBench, while better preserving non-edited regions. The method delivers up to 1.8$\times$ speedup at 1024$^2$ resolution and 2$\times$ speedup at 2048$^2$ resolution. The code will also be released publicly.

Neural network training can be viewed as a discrete-time dynamical system, suggesting that its optimisation trajectory can be learned and extrapolated to reduce the cost of repeated backpropagation. Motivated by Koopman operator theory, which represents nonlinear dynamics as linear evolution in a lifted space, we propose a training acceleration method based on augmented Koopman dynamics. Instead of approximating the evolution of parameters alone, our method lifts the training state to include both model parameters and optimiser-dependent internal states, enabling the learned operator to better capture the dynamics of modern optimisers with momentum or adaptive gradient statistics. We estimate a strided, multi-step Koopman operator from training snapshots and use matrix-vector multiplications to extrapolate future parameters and optimiser states, effectively bypassing an adaptive number of backpropagation steps. This formulation suppresses high-frequency stochastic fluctuations induced by mini-batch training and reduces the storage required for operator estimation. To improve robustness, we further introduce a safeguarding mechanism that prevents Koopman-extrapolated states from degrading performance relative to the most recent backpropagation iterate. The method can be integrated with standard optimisers, including Adam, AdamW, SGD with momentum, Adadelta, and other commonly used variants. Experiments across multiple optimisers show that the proposed approach significantly reduces both the number of backpropagation steps and the wall-clock training time required to reach a target accuracy, while maintaining, and in many cases improving, final model performance.

Online bilevel optimization (OBO) has emerged as a powerful framework for many machine learning problems. Prior works have developed several algorithms that minimize the standard bilevel local regret or the window-averaged bilevel local regret of the OBO problem, but the optimality of existing regret bounds remains unclear. In this work, we establish optimal regret bounds for both settings. For standard bilevel local regret, we propose an algorithm with adaptive iteration strategy that achieves the optimal regret $\Omega(1+V_T)$ with at most $O(T\log T)$ total inner-level gradient evaluations. We further develop a fully single-loop algorithm whose regret bound includes an additional gradient-variation terms. For the window-averaged bilevel local regret, we design an algorithm that captures linear environmental variation through a novel window-based analysis and achieves the optimal regret $\Omega(T/W^2)$. The algorithm also supports an efficient single-loop structure, achieving an $O(T/W)$ regret bound with $O(WT)$ total gradient evaluations. Experiments validate our theoretical findings and demonstrate the practical effectiveness of the proposed methods.

Submodular functions have applications in several domains, \eg, text, vision, speech, \etc. Recent works have proposed neural architectures that are submodular functions by design. However, they do not explicitly capture pairwise interactions between set elements. In this work, we begin with the characterization of pairwise submodular functions, which capture the pairwise interaction across the elements of the input set. We observe that pairwise interactions yield monotone supermodular functions, since the number of terms grows quadratically in terms of the input set size, which poses a significant challenge in converting it into a submodular function. To address it, we provide a novel result that a derivative rate-controlled concave function can transform a monotone supermodular function into a monotone submodular function. We also extend our results to monotone $\alpha$-submodular functions. Leveraging this characterization, we design multi-layer neural cross-interaction architectures for monotone submodular functions and analyze their expressivity. Finally, we perform several experiments which show that our model performs better than existing baselines.

Learning from positive and unlabeled data (PU learning) is a weakly supervised variant of binary classification in which the learner receives labels only for (some) positively labeled instances, while all other examples remain unlabeled. Motivated by applications such as advertising and anomaly detection, we study active PU learning, where the learner adaptively queries instances from an unlabeled pool, but a label is revealed only when the queried instance is positive and an independent coin flip succeeds; otherwise the learner receives no information. This paper provides the first theoretical analysis of the label complexity of active PU learning.

Continuous-time recurrent models update hidden states as observations arrive, making them suitable for irregular and non-stationary temporal data. Liquid Neural Networks (LNNs) extend this formulation through input-dependent state dynamics, but they still rely on a single evolving state to integrate new observations and retain earlier context. Under long temporal gaps or missing inputs, earlier evidence can weaken as the hidden state continues to evolve. We propose Active Memory Feedback Loop (AMFL), a memory-augmented LNN architecture that feeds associative retrieval back into the recurrent computation. AMFL combines a learned Short-Term Memory (STM) that stores global sequence prototypes with a sequence-local Iconic Memory (IM) that adapts through novelty-gated, gradient-free updates. After each liquid refinement step, IM retrieves an associative representation, forms a residual correction, and writes the correction into a feedback buffer that conditions the following recurrent update. Memory therefore influences the evolving liquid representation rather than only augmenting the final prediction head. We evaluate AMFL on seven time-series benchmarks covering classification, regression, continual learning, and robustness to missing inputs. AMFL obtains the best result on six of seven benchmarks and remains competitive on the remaining task. Ablations show that the gains arise from looped memory feedback, controlled IM adaptation, and residual correction. Additional missing-input and long-sequence forecasting experiments further support the role of associative feedback in preserving task-relevant context under partial observability.

Offline goal-conditioned reinforcement learning (GCRL) aims to learn a single policy that adapts its behavior to arbitrary goals from a fixed dataset, making injection of the goal information into the policy a central design question. While recent works increasingly leverage expressive conditional generative modeling architectures, which have demonstrated remarkable success in computer vision, architectural choices for goal conditioning in offline GCRL remain under-explored. We identify that goals in GCRL differ from typical conditioning signals in two ways: they are often noisy or redundant, and they are consumed within a time-sequential episode where the relevant context of a fixed goal shifts as the agent's state evolves. Motivated by these observations, we propose State-Anchored Goal Conditioning (SAGC), a modulation-based goal conditioning scheme that derives its scale and shift parameters by jointly processing state and goal representations in a learned latent space. We further introduce AdaptFlow, an end-to-end trainable offline GCRL architecture that integrates conditional flow matching with SAGC. AdaptFlow outperforms matches all baselines on 23 out of 27 OGBench tasks.

Cooperative multi-agent reinforcement learning (MARL) has long faced scalability challenges due to the exponential growth of the state-action space as the number of agents increases. Existing methods typically enhance scalability by filtering out communications with low-relevance agents. However, such filtering often relies on global state and fixed graph distribution, limiting adaptability in large-scale, communication-constrained environments. In this paper, we propose a scalable MARL method, called Adaptive Communication Range PPO (ACR-PPO), that decomposes the decision-making under communication budget constraints as a sequential process: a communication policy first selects each agent’s communication range within a given budget, followed by a behavior policy that conditions actions on the resulting neighborhood observations. More importantly, we provide a theoretical guarantee of monotonic performance improvement under communication budget constraints. Experiments across diverse scenarios demonstrate that ACR-PPO preserves policy performance while significantly reducing communication costs through adaptive range control.

Personalizing Large Language Models (LLMs) directly on local edge servers is becoming increasingly important for privacy-preserving and context-aware applications. However, this potential is bottlenecked by hardware resources: limited GPU memory permits only a fixed number of train-ready LoRA adapters, while compute constraints enforce sequential fine-tuning updates. This creates a critical challenge: scheduling scarce update opportunities across a dynamic stream of resident tenants to improve overall model quality and scheduling stability. Crucially, standard Multi-Armed Bandit (MAB) algorithms fail to distinguish between low-potential tenant streams and fully saturated adapters, leading to wasteful exploration on tasks that offer little further gain. To address this, we propose WCA-UCB (Windowed Convergence-Aware UCB), a scheduler designed for this piecewise-converging non-stationary environment. By formally modeling fine-tuning as a slot-constrained bandit problem with piecewise-converging rewards, WCA-UCB detects when an adapter has saturated to pause its training and resets stale statistics after tenant replacement or harmful drift. We prove a dynamic regret bound of $\tilde{O}(\log T)$ and validate the system on an edge-like single-GPU LoRA prototype running Qwen2.5-1.5B. Results demonstrate that WCA-UCB reduces model perplexity by $2.6$% to $8.6$% and uses up to $5.8\times$ fewer training-target switches than strong non-stationary baselines. These results highlight the necessity of convergence-aware scheduling for scalable local LLM personalization.

Gaussian process (GP) bandits provide a powerful framework for performing blackbox optimization of unknown functions. The characteristics of the unknown function depend heavily on the assumed GP prior. Most work in the literature assume that this prior is known but in practice this seldom holds. Instead, practitioners often rely on maximum likelihood estimation to select the hyperparameters of the prior - which lacks theoretical guarantees. In this work, we study two algorithms for joint prior selection and regret minimization in GP bandits based on GP Thompson sampling (GP-TS): Prior-Elimination GP-TS (PE-GP-TS) that disqualifies priors with poor predictive performance, and HyperPrior GP-TS (HP-GP-TS) that utilizes a bi-level Thompson sampling scheme. We theoretically analyze the algorithms and establish a sublinear regret bound for HP-GP-TS. In addition, we demonstrate the effectiveness of these algorithms compared to the alternatives through extensive experiments with synthetic and real-world data.

Speculative decoding accelerates generative inference of large language models (LLMs) by using a small draft model to propose multiple candidate tokens, which are then verified in parallel with the target model in a single decoding iteration. While the state-of-the-art method of tree-based speculative decoding helps improve generation throughput over non-speculative inference, deploying it in LLM serving systems often yields suboptimal performance due to dynamically changing serving conditions. In particular, our analysis shows that the optimal tree configuration---the one that maximizes performance---varies with two key serving conditions: request rate and per-request characteristics. Based on the analysis, we present AdaTree, a plug-in component for LLM serving systems that dynamically adapts tree configurations to varying serving conditions. AdaTree predicts model execution time and acceptance length across different tree configurations and selects the one that maximizes speculative decoding efficiency. To capture the non-linear relationship between tree configuration and model execution time, AdaTree employs decision trees as its core modeling primitive. Our evaluation shows that AdaTree consistently outperforms both chain-based and static tree-based speculative decoding across diverse serving conditions.

Temporally extended exploration via graph Laplacian-based options is a promising approach to sparse-reward reinforcement learning (RL), but existing methods either do not explicitly target novelty or fail to scale to pixel-based domains under function approximation. Novel Exploration via Orthogonality (NEO) addresses the first issue by constructing options that navigate from highly visited regions toward less visited ones, yet prior results were limited to settings where exact eigenvectors can be computed. We present a scalable extension of NEO to pixel-based domains, built on three contributions. First, we use a novelty-weighted continuous Laplacian graph-drawing objective, which enables RL with continuous observations. Second, we embed the resulting eigen-potential options within a hierarchical reinforcement learning framework, enabling coherent temporally extended behavior. Third, we observe that learned eigen-potential rewards are directional but locally unreliable under online approximation; we therefore augment each option reward with a novelty bonus, a novel design idea that proves essential for stabilizing option learning while preserving novelty-directed exploration. Together, these contributions yield stronger and more persistent exploration, enabling longer option rollouts and better access to hard-to-reach novel states. Empirically, our method significantly outperforms both the prior scalable Laplacian-option baseline and a direct extension of NEO on sparse-reward benchmarks under a fixed budget. On Montezuma's Revenge, our best variant achieves approximately 1.8x higher return than both baselines. On Venture, both baselines yield returns near zero, whereas our method achieves a return of 1135. Across seven hard ProcGen games, our method achieves approximately 3.5x and 5.6x higher aggregate normalized return than the two baselines, respectively.

3D scene graph prediction is commonly supervised with local object and predicate classification losses. Although effective for slot-wise label prediction, this formulation does not provide a graph-level semantic distance between a predicted scene graph and its ground truth: semantically mild label errors and structurally disruptive relational errors can be treated similarly, and subject-predicate-object facts are not compared as scene-level units. We propose Contextual Hellinger Triplet Geometry, a graph-level supervision objective that measures the semantic and structural discrepancy between 3D scene graphs. Our key idea is to reinterpret the conventional classifier outputs of a 3D scene graph predictor as an aligned probabilistic scene graph field, where object and predicate predictions jointly define contextual subject-predicate-object facts. This formulation yields a differentiable metric-based training objective with an efficient factorized implementation and can be applied to existing 3D scene graph prediction architectures without modifying their network design. Experiments on 3DSSG demonstrate that our supervision improves downstream graph-conditioned 3D scene generation, while ablation studies confirm the complementary roles of the proposed graph-level terms and user studies show better graph descriptiveness, relation plausibility, and structural preservation.

Federated learning (FL) has become a foundational paradigm for multi-institutional medical AI, allowing hospitals and research centers to jointly train diagnostic models without exchanging patient records. This privacy promise, however, is increasingly contested: a malicious or honest-but-curious server can launch model inversion attacks (MIAs) that reconstruct private patient images directly from shared model updates, and recent scalable, closed-form attacks penetrate even secure aggregation at clinically realistic batch sizes. Existing defenses face an unsatisfactory dilemma. Gradient-perturbation methods such as differential privacy and pruning trade away the diagnostic accuracy on which clinical reliability depends, while cryptographic protocols add system complexity yet still leave updates exposed to these scalable attacks. We propose Aegis, a principled client-side defense that breaks this dilemma without perturbing patient data or modifying the FL protocol. Our key insight is that the success of every known MIA is fundamentally bounded by the local batch size relative to the model's leakage capacity; once this limit is exceeded, distinct samples collide and reconstructions collapse into indistinguishable mixtures. Aegis turns this universal bottleneck into a defense: each client superimposes onto its real update a masking gradient computed on locally synthesized, task-relevant data, deliberately pushing the effective batch beyond the attack's recovery capacity. We complement the design with theoretical convergence guarantees under standard convex assumptions and evaluate Aegis on MNIST, CIFAR-10, and three MedMNIST modalities (chest X-ray, abdominal CT, colon pathology). Aegis neutralizes three state-of-the-art MIAs while preserving model utility and incurring only modest overhead, offering a practical privacy primitive for medical FL.

LLM-based agents are increasingly deployed in real-world applications through tool-use APIs, yet training them for specific environments remains fundamentally difficult: real-world applications provide no pre-defined tasks or verifiers, no faithful simulators, and limited budget for large-scale environment interaction. In this paper, we propose AgentBrew, an offline training framework that learns effective tool-use policies from a single batch of raw interaction trajectories, without task verifiers or iterative on-policy rollouts. The agent first explores the target environment to collect a raw trajectory corpus without quality filtering. To extract training signal from this noisy corpus, retrospective task inference reconstructs an aligned instruction for each trajectory based on its actual outcome, and PMI-Based credit assignment decomposes the trajectory's total information about the inferred instruction into additive per-action credits via pointwise mutual information (PMI). These credits weight the policy training objective, amplifying informative actions while suppressing ineffective ones. On three real-world MCP applications (GitHub, Notion, PostgreSQL), AgentBrew improves Qwen3-32B by +8.7 Acc / +9.7 Score on average, surpassing Qwen3-235B (+2.3 / +4.4) and outperforming rejection sampling (+5.9 / +10.3). These results demonstrate that fine-grained offline learning can recover useful supervision from raw trajectories that filtering-based approaches would discard.

LLM-based multi-agent systems are increasingly deployed on long-horizon tasks, but a single decisive error is often accepted by downstream agents and cascades into trajectory-level failure. Existing work frames this as \emph{post-hoc failure attribution}, diagnosing the responsible agent and step after the trajectory has ended. However, this paradigm forfeits any opportunity to intervene while trajectory is still unfolding. In this work, we introduce AgentForesight, a framework that reframes this problem as online auditing: at each step of an unfolding trajectory, an auditor observes only the current prefix and must either continue the run or alarm at the earliest decisive error, without access to future steps. To this end, we curate AFTraj-2K, a corpus of agentic trajectories across Coding, Math, and Agentic domains, in which safe trajectories are retained under a strict curation pipeline and unsafe trajectories are annotated at the step of their decisive error via consensus among multiple LLM judges. Built on that, we develop AgentForesight-7B, a compact online auditor trained with a coarse-to-fine reinforcement learning recipe that first equips it with a risk-anticipation prior at the failure boundary on adjacent safe/unsafe prefix pairs, then sharpens this prior into precise step-level localization under a three-axis reward jointly targeting the what, where, and who of an audit verdict. Across AFTraj-2K and an external Who\&When benchmark, AgentForesight-7B outperforms leading proprietary models, including GPT-4.1 and DeepSeek-V4-Pro, achieving up to +19.9\% performance gain and 3$\times$ lower step localization error, opening the loop from post-hoc failure detection to enabling deployment-time intervention.

LLM agents are expected to act over multiple turns, using search, browsing interfaces, and terminal tools to complete user goals. Yet not every goal is well specified or achievable in the available environment. In such cases, a reliable agent should recognize that further interaction is unlikely to help and abstain from additional tool calls. We define Agentic Abstention, the problem of deciding when an agent should stop acting under uncertainty. Unlike standard LLM abstention, which is usually evaluated as a single-turn answer-or-abstain decision, agentic abstention is a sequential decision problem: an agent can answer, abstain, or gather more information at each turn, and the need to abstain may only become clear after interacting with the environment. We study this problem across web shopping, terminal environments, and question answering, evaluating 13 LLM-as-agent systems and 2 agent scaffolds on more than 28,000 tasks. Our results show that the main challenge is not only whether agents can abstain, but also when they abstain. Some agents never abstain when they should, while others do so only after many unnecessary interactions. This gap is especially large on tasks where the instruction appears feasible until the environment reveals otherwise (e.g., no valid result matches the instruction). We further find that model scale, reasoning, and agent scaffolding affect abstention in different ways, with larger or more capable models not always performing better at timely abstention. Finally, we introduce CONVOLVE, a context engineering method for improving agentic abstention that distills full interaction trajectories into reusable stopping rules. On WebShop, CONVOLVE substantially improves timely abstention without updating model parameters, raising Llama-3.3-70B's timely recall rate from 26.7 to 57.4. % , and overall recall from 83.2 to 100.0. Our dataset and code are available at https://anonymous.4open.science/r/agentic-abstention-A908

Recent advances in Large Language Models (LLMs) have enabled agentic systems capable of solving complex tasks through multi-turn planning, tool use, verification, and memory updates. However, learning agentic systems remains difficult due to two fundamental challenges, i.e., (1) long-horizon credit assignment, where supervision is available only at the final outcome, and (2) imbalanced data distributions, where dominant data patterns bias optimization and weaken adaptation to rare but informative reasoning behaviors. In this paper, we propose Fair Multi-Level Preference Optimization (Fair-MPO or $\Phi$-MPO), a new preference optimization framework for agentic learning. We first show that Multi-Level Preference Optimization provides a principled and more computationally efficient framework for long-horizon reasoning. Then, we introduce a Fair Multi-Level Objective that addresses imbalance in agentic learning. We provide a comprehensive theoretical analysis demonstrating that our approach addresses both long-horizon reasoning and data imbalance. Our experiments on agentic reasoning benchmarks demonstrate that our approach achieves State-of-the-Art (SOTA) performance.

Neural architecture search (NAS) methods have grown increasingly efficient, yet they remain bounded by manually engineered search spaces that require substantial domain expertise and must be rebuilt for every new task. Large language models (LLMs) can generate architectures in an open-ended space, but how to optimally divide labor between LLM-driven design and NAS-driven search remains unexplored. We propose a mechanism that bridges these two paradigms: an LLM produces a high-quality seed architecture, then decomposes it into a slotted architecture---a scaffold with named, interchangeable module slots that automatically defines a bounded, task-specific search space for conventional NAS to explore, without manual engineering. We instantiate this mechanism in AgentNAS, a modular three-phase pipeline in which each component's contribution can be measured independently. On 17 tasks spanning classification, dense regression, segmentation, and multi-label tagging across diverse modalities (NAS-Bench-360 and NAS-Unseen), AgentNAS establishes a new state of the art on 11 tasks, outperforming published baselines including task-specific expert designs. Ablation studies show that the two search mechanisms are broadly complementary: the LLM-generated seed already surpasses published baselines on the majority of tasks, and NAS delivers additional gains in most cases through combinatorial recombination across slots---a mode of search that independent LLM samples cannot replicate. These patterns hold across three LLMs of different capability levels, confirming that the division of labor is robust.

Evaluation of software engineering (SWE) agents is dominated by a binary signal: whether the final patch passes the tests. This outcome-only view treats a principled solution and a chaotic trial-and-error process as equivalent. We show that this equivalence is empirically false. We evaluate 2,614 OpenHands trajectories from eight model backends on SWE-bench Verified. Of the 60 tasks in this corpus, 47 have enough passing trajectories to construct task-level process references, yielding a 1,815-trajectory evaluation subset. Among passing trajectories in this subset, 10.7% exhibit behavior we call a Lucky Pass: regression cycles, blind retries, missing verification, or temporally disordered exploration, implementation, and verification. We introduce AgentLens, a framework for process-level assessment of SWE-agent trajectories, and release AgentLens-Bench, a dataset of 1,815 trajectories annotated with quality scores, waste signals, divergence points, and 47 task-level Prefix Tree Acceptor (PTA) references. AgentLens combines two components. First, it merges multiple passing solutions for the same task into a PTA reference space of correct behaviors. Second, it uses a context-sensitive intent-stage labeler that assigns actions to Exploration, Implementation, Verification, or Orchestration using trajectory history rather than tool identity alone. On AgentLens-Bench, the composite score separates passing trajectories into Lucky, Solid, and Ideal tiers; decomposes Lucky Passes into five recurring mechanisms; and changes how the eight evaluated model backends are ranked compared with pass rate alone. Across these models, AgentLens classifies between 0.5% and 23.2% of successful trajectories as Lucky, and some models move by as many as five rank positions when ranked by quality score instead of pass rate. We release the anonymized project repository, including the AgentLens-Bench dataset and AgentLens SDK, at https://anonymous.4open.science/r/agentlens-app-6810/.

Agent applications are evolving into workflows composed of large language model (LLM) calls and tool invocations, termed modules. Existing distributed agent systems typically adopt a disaggregated architecture, deploying different modules on separate devices. While this enables pipeline parallelism, imbalanced module latencies introduce severe pipeline bubbles that limit throughput. A natural alternative is an aggregated architecture that co-deploys all modules across all devices, eliminating bubbles by keeping every device continuously active. However, this introduces request blocking: short requests are forced to progress in lockstep with longer ones in the same batch, and newly arriving requests must wait until the entire ongoing batch completes the full workflow. To address this, we propose AgentWeave, an efficient distributed serving system for agent applications built on the aggregated architecture. Its core mechanism, flow decomposition, decomposes long agent requests into fine-grained execution units while preserving agent semantics, allowing short and newly arriving requests to proceed without being stalled by long-running ones. Complementing this, a flow-adaptive scheduler dynamically balances the throughput gains of decomposition against its scheduling overhead. Experiments across four representative agent applications demonstrate that AgentWeave improves throughput by $1.81\times$, reduces average latency by $2.45\times$, and reduces P90 latency by $2.04\times$ over state-of-the-art baselines.

Modern agricultural data is sourced from diverse platforms and spans multiple spatial scales, ranging from ground-level close-up photography to Unmanned Aerial Vehicle (UAV) aerial observation and satellite remote sensing imagery. Accordingly, agricultural multimodal reasoning demands robust cross-scale spatial understanding. However, due to the lack of multi-view agricultural benchmark datasets, existing multimodal large language models (MLLMs) exhibit severe ground-level bias, which leads to scale confusion then semantic collapse in agricultural perception tasks, such as misinterpreting farmland imagery as walls or floors. To address this, we introduce \textbf{AgroOmni}, a large-scale multi-view training corpus with 288K Visual Question Answering pairs covering 56 specialized task categories across 14 task types, designed to capture diverse scales in modern precision agriculture. Built on this dataset, we propose AgroNVILA, which achieves a new state-of-the-art of 62.32\% on the AgroMind benchmark ($+15.03\%$ over GPT-5.2), effectively mitigating the multi-view cross-scale gap for holistic agricultural understanding. Diagnostic evaluations on AgMMU further reveal an inherent heterogeneity between macro-priors and micro-diagnostics through constrained zero-shot performance. Meanwhile, even minimal fine-tuning leads to a dramatic performance gain of AgroNVILA on AgMMU, strongly demonstrating its generalization capability empowered by AgroOmni. Full training scripts are publicly available at \url{https://anonymous.4open.science/r/AgroOmni-6510}.

Representation alignment has recently emerged as an effective paradigm for accelerating Diffusion Transformer training. Despite their success, existing alignment methods typically impose a fixed supervision target or a fixed alignment granularity throughout the entire denoising trajectory, whether the guidance is provided by external vision encoders, internal self-representations, or VAE-derived features. We argue that such timestep-agnostic alignment is suboptimal because the useful granularity of representation supervision changes systematically with the signal-to-noise ratio. In high-noise regimes, diffusion models benefit more from coarse semantic and layout-level anchoring, whereas in low-noise regimes, the training signal should emphasize spatially detailed and structurally faithful refinement. This non-stationary alignment behavior creates a representational mismatch for static single-level supervisors. To address this issue, we propose Adaptive Hierarchical Prior Alignment (AHPA), a lightweight alignment framework that exploits the hierarchical representations naturally embedded in the frozen VAE encoder. Instead of using only a single compressed latent as the alignment target, AHPA extracts multi-level VAE features that provide complementary priors ranging from local geometry and spatial topology to coarse semantic layout. A timestep-conditioned Dynamic Router adaptively selects and weights these hierarchical priors along the denoising trajectory, thereby synchronizing the alignment granularity with the model's evolving training needs. Extensive experiments show that AHPA improves convergence and generation quality over baselines and incurs no additional inference cost while avoiding external encoder supervision during training.

Federated fine-tuning of large language models (LLMs) under heterogeneous client capacities is increasingly important, and low-rank adaptation (LoRA) with heterogeneous ranks makes it practical. Existing server-side aggregators compose local matrix updates $\Delta W_i$ from client-specific LoRA factors $(A_i,B_i)$ before aggregating them. Although these matrix updates lie in the same weight space $\mathbb{R}^{m\times n}$ and can be averaged directly, the LoRA factors that generate them may expressed client-specific local basis. In the typical compose--aggregate--factorize--dispatch loop, this basis inconsistency can make the rank-$r_i$ reference prefix dispatched to lower-rank clients unstable after server-side refactorization. In this work, we propose FedAbA, a basis-consistent federated LoRA aggregation framework for heterogeneous ranks. FedAbA first extracts each client's local basis, aligns it to the corresponding prefix of the global reference basis, and constructs a basis-consistent reconstruction of each local matrix update before aggregation. It then aggregates these reconstructions with weights that combine client data size and consistency score, followed by exact coefficient-space SVD refactorization and gauge fixing before rank-prefix dispatch. Our theoretical analysis clarifies the role of basis consistency and supports the proposed consistency-aware aggregation mechanism. Extensive experiments show that FedAbA outperforms representative federated LoRA baselines, with the largest gains under the strongest rank heterogeneity.

Generative policies based on expressive model classes, such as diffusion and flow matching, are well-suited to complex control problems with highly multimodal action distributions. Their expressivity, however, comes at a significant inference cost: generating each action typically requires simulating many steps of the generative process, compounding latency across sequential decision-making rollouts. We introduce flow map policies, a novel class of generative policies designed for fast action generation by learning to take arbitrary-size jumps---including one-step jumps---across the generative dynamics of existing flow-based policies. We instantiate flow map policies for offline-to-online reinforcement learning and formulate online adaptation as a trust-region optimization problem that improves the critic's $Q$-value while remaining close to the offline policy. We theoretically derive Flow Map $Q$-Guidance Training (FMQ), a principled closed-form learning target that is optimal for adapting offline flow map policies under a critic-guided trust-region constraint. We further introduce $Q$-Guided Beam Search (QGBS), a stochastic flow-map sampler that combines renoising with beam search to enable iterative inference-time refinement. Across $12$ challenging robotic manipulation and locomotion tasks from OGBench and RoboMimic, FMQ achieves state-of-the-art performance in offline-to-online RL, outperforming the previous one-step policy MVP by a relative improvement of $21.3$% on the average success rate.

Recommender systems (RS) continually learn from new user-item interactions while addressing heterogeneous deletion requests. Existing continual recommendation methods focus on retention but do not erase targeted interactions, whereas recommendation unlearning methods are largely offline and do not ensure stability after future learning updates. This creates a critical failure mode: deleted signals can be reactivated through shared user--item representations, which we call collaborative signal regrowth. To address this, we introduce ALIGN-Rec, a model-agnostic online framework for continual recommendation under interleaved learn/unlearn requests. Specifically, ALIGN-Rec tracks low-rank curvature surrogates to identify retain and forget subspaces, constructs compact retain/forget summaries, and performs geometry-aware updates that preserve retained utility while suppressing deleted-signal directions. We instantiate ALIGN-Rec with CARE, an efficient optimizer based on randomized curvature sketching and adaptive summary selection. We provide theoretical guarantees for geometry tracking, summary approximation, and directional suppression in the forget subspace. Extensive experiments across datasets, deletion granularities, and stream settings demonstrate effectiveness aligning with our theoretical guarantees. The code and implementation are available at https://anonymous.4open.science/r/alignrec-FFD8.

Protein binder design has largely optimized for affinity alone, leaving conformational selectivity unaddressed: for allosteric targets such as kinases, nuclear receptors, and GPCRs, a binder that engages both active and inactive states provides no functional specificity regardless of how tightly it binds. We introduce **AlloGen**, a modular framework that decouples backbone generation from a learned state-selectivity scorer $Q_\theta$, an SE(3)-invariant interface graph transformer trained via a two-phase curriculum that first grounds interface geometry before imposing conformational discrimination. Because $Q_\theta$ is fully differentiable and generator-agnostic, it integrates with any backbone generator as a passive reranker or an active gradient-based guide without retraining. Trained on 65 targets spanning 15 protein families, $Q_\theta$ generalizes to held-out out-of-distribution targets where energy-based baselines fail entirely, and all 15 evaluated generator--guidance combinations achieve positive conformational selectivity averaged over the held-out targets, with the best reaching $\bar{S}=+0.677$. Our anonymous code repository can be found at https://anonymous.4open.science/r/AlloGen_NeurIPS-04CB.

Recaptioned image-text corpora are now standard for text-to-image (T2I) training, with VLM captioners replacing sparse alt-text by dense descriptions. Yet a recaptioned corpus is not only a collection of captions: it is a supervision distribution induced by a documented captioning policy ($ \pi $), captioner ($ V_c $ ), and source corpus ($ C $). Length-correlated proxies miss caption-register artifacts and downstream T2I benchmarks entangle the corpus with training choices, so this distribution is hard to audit at corpus scale. We introduce a reusable matched-budget audit framework for recaptioned supervision distributions $ D_{\pi,V_c,C} $: at a fixed $ B = 64 $ lexical-unit window it reports a five-axis profile spanning prompt-side coverage, image-conditioned faithfulness, and caption-surface health, with claimed controllable basic units (CBU) as the shared denominator. We instantiate the framework on seven paired comparisons over five public source corpora. Across the four cross-corpus pairs, the released surface raises supported CBU per caption by +3.40 to +6.35 under both Qwen and Gemma Judges, and on CC12M the same framework exposes a long-vs-dense frontier that is consistent across both judges and four budgets. We release the audited multi-source recap corpus together with the audit-artifact bundle.

Humans, animals, and modern machine learning models exhibit impressive abilities to learn complex behaviors and generalize these behaviors to unseen situations. This ability requires us to learn rules and regularities that allow for such generalizations. At the same time, in most complex environments, any rule will have its exceptions. How do learning systems balance between learning general regularities and memorizing exceptions? We argue that a lack of task paradigms has hindered the study of this essential ability. To address this gap, we introduce a novel task, transitive inference with exceptions, that tests for relational generalization and memorization of an exception to the relational rule. We then analytically characterize the behavior of a simple, theoretically tractable model of neural network learning (kernel ridge regression) across a broad family of representations and task parameters. We find that these models can balance between relational generalization and memorization, but unlike for transitive inference without an exception, successful generalization is sensitive to the specific representational geometry. We explain why this task is more challenging mechanistically by drawing on our analytical theory. Finally, we validate our theoretical insights in pretrained language models that are finetuned on ordered relations, finding that these models successfully generalize according to the transitive rule, but also make the kinds of systematic mistakes predicted by our theory. Overall, our theory shows how learning systems can balance between relational generalization and memorization, explains how this can go wrong, and emphasizes the need for new task paradigms designed to probe this ability.

The evolution of mathematics is shaped importantly by interestingness: researchers choose which problems to pursue, and students choose which problems to engage with, based on expectations of interest and challenge. As AI systems, particularly large language models (LLMs) that operate flexibly over natural language and formal mathematics, are increasingly used in mathematics research and education, it becomes crucial to characterize how closely their judgments align with people from different mathematical backgrounds. We study whether LLMs align with human interestingness judgments by comparing LLM ratings with those of two populations, crowdsourced participants with college math experience and International Math Olympiad competitors. Although many LLMs broadly agree with human notions of interestingness, they largely fail to match the distribution of human judgments. They also weakly align with why humans find problems interesting, with low correlation to human-selected rationales. Finally, we evaluate LLMs’ ability to generate interesting problems and find that, after filtering for validity, LLMs are able to generate engaging problems. We conclude with takeaways, including the need for multi-LLM human-AI collaborative systems, that highlight both the promise and current limits of LLMs as partners in mathematical reasoning.

Motivated by the challenge of stabilizing unknown linear dynamical systems (LDS) from observations, we study the fundamental prerequisite of online prediction. Our goal is to achieve sublinear regret with a memory footprint that adapts to the intrinsic complexity of the dynamics rather than the full hidden-state dimension. We focus on the practically central regime of systems with low *instability complexity*—eigenvalues outside the real stable interval that do not decay rapidly, together with non-semisimple modes—potentially embedded in an otherwise stable real spectrum of much higher dimension; we write $k$ for this count. This regime is the primary setting in which stabilization is plausible: we show that many systems with high instability complexity cannot be stabilized without exponentially large controls. Thus, prediction is meaningful for stabilization precisely when the instability complexity is small. Within this regime, we introduce a unified online algorithm that handles every LDS (including systems with complex or exploding modes) with a learnable parameter count of $\widetilde{O}(k^2)$, completely independent of the number of stable real modes. Finally, we show that any bounded-coefficient filter-based predictor requires at least $k$ filter directions.

Vision Transformers (ViTs) achieve strong performance across vision tasks, yet their deployment with low-precision early exiting remains fragile. Existing quantization methods assume static full-depth execution, making them unstable when exit decisions are perturbed by quantization noise, which can amplify errors along dynamic inference paths. In this paper, we introduce Amortized-Precision Quantization (APQ), a utilization-aware formulation that accounts for layer-wise stochastic exposure to quantization noise and reveals depth--precision trade-offs. Building on APQ, we propose Mutual Adaptive Quantization with Early Exiting (MAQEE), a bi-level framework that jointly optimizes exit thresholds and bit-widths under explicit risk control to improve inference stability. MAQEE establishes a superior Pareto frontier in the accuracy--efficiency trade-off, reducing BOPs by up to 95% while maintaining accuracy and outperforming strong baselines by up to 20% across classification, detection, and segmentation tasks. Source code is available at https://anonymous.4open.science/r/MAQEE-5F2F.

Recovering a signal from its degraded measurements is a long standing challenge in science and engineering. Recently, zero-shot diffusion based methods have been proposed for such inverse problems, offering a posterior sampling based solution that leverages prior knowledge. Such algorithms incorporate the observations through inference, often leaning on manual tuning and heuristics. In this work we propose a rigorous analysis of these approximate posterior samplers, relying on a Gaussianity assumption of the prior. Under this regime, we show that both the ideal posterior sampler and diffusion-based reconstruction algorithms can be expressed in closed-form, enabling their thorough analysis and comparisons in the spectral domain. Building on these representations, we introduce a principled framework for parameter design, replacing heuristic selection strategies used to date. The proposed approach is method-agnostic and yields tailored parameter choices that jointly account for the characteristics of the prior, the degraded signal, and the diffusion dynamics. We show that our spectral recommendations differ structurally from standard heuristics and vary with the diffusion step size, resulting in a consistent balance between perceptual quality and signal fidelity.

Multimodal large language models (MLLMs) achieve impressive performance on audio-visual question answering, but correct answers do not necessarily imply perceptually grounded reasoning. Models may exploit linguistic priors or dataset shortcuts without attending to the visual regions and audio events that actually justify their predictions. Existing benchmarks predominantly evaluate final answer accuracy, providing limited insight into whether models truly “see” or “hear” the evidence required for reasoning. We introduce ANCHOR, a benchmark and evaluation framework for auditing perceptual grounding in multimodal reasoning. ANCHOR requires models to produce grounded chain-of-thought explanations that explicitly reference audio-visual evidence, including representative frames, spatial bounding boxes, and salient audio cues. To build a reliable gold standard, we combine automated annotation generation with a human-in-the-loop refinement pipeline, producing 224 human-verified videos and 975 question–answer pairs with aligned reasoning and localization annotations from 2.3K videos and 8.2K QA pairs. We further propose a metric and evaluation protocol that jointly measure answer correctness and reasoning-grounding alignment, complemented by multi-judge scoring of hallucination, logical divergence, final answer, and a token-based conciseness measurement. Experiments on nine state-of-the-art MLLMs reveal a substantial faithfulness gap: models with strong answer performance often produce explanations that are only weakly grounded in the underlying perceptual evidence. Ablation results confirm the importance of explicit grounding cues: removing bounding boxes reduces accuracy by 7.7 points, and performance drops from 57.7% with full cues to 16.1% in the audio-only setting. These findings show that answer-only evaluation is insufficient for multimodal reasoning and position ANCHOR as a new testbed for developing systems that are accurate, interpretable, and trustworthy.

Remote photoplethysmography (rPPG) enables non-contact physiological measurement from facial videos, but performance often degrades under unseen environments due to severe domain shifts. Continual learning offers a practical paradigm for updating rPPG models over sequential domains. However, it is challenged by two coupled issues: First, adapting to new domains may overwrite previously acquired physiological representations, resulting in the degradation of physiological consistency across domains. Second, drastic environmental variations induce heterogeneous facial video distributions, making the adaptation process unstable and prone to suboptimal convergence. To address these challenges, we propose ApexPhys, a continual rPPG framework that simultaneously anchors physiological invariance and expands domain plasticity. Specifically, we introduce a Backbone Consistency Representation Preservation (BCRP) mechanism, which leverages the Fisher Information Matrix to identify and preserve parameters critical to physiological consistency. To enhance adaptation flexibility under diverse environmental shifts, we further propose a Hierarchical Expansion Strategy (HES) that autonomously perceives layer-wise distribution discrepancies and dynamically expands hierarchical adaptation branches. Additionally, we develop a Prior-Guided Pathfinding (PGP) strategy, which leverages parameter inheritance to guide the adaptation along a stable trajectory. Experiments on five public datasets demonstrate that ApexPhys achieves strong anti-forgetting performance and robust adaptation under continual learning.

Adversarial attacks optimized on a single open-weight LLM can transfer to and jailbreak architecturally different models, allowing an attacker with white-box access to one model to compromise independently deployed systems. This creates a shared vulnerability across models, yet existing defenses are not designed for this cross-model threat. We find that transfer aligns with shared internal representation geometry, making it a natural defense target. We find that cross-model transfer aligns with shared internal representation geometry, making it a natural defense target. \textbf{AnchorRep} targets this geometry directly with a lightweight LoRA adapter that pushes the defended model's internal representations of harmful prompts away from those of a frozen anchor model on the same prompts. Training uses a small set of harmful prompts and no adversarial examples. Across five models and four architectural families, AnchorRep reduces cross-model attack success rate to $\leq$1.1\% on 2{,}000 transferred attacks (0\% on two), including the largest drop on Mistral ($36\% \to 1.1\%$). Existing defenses can reduce transfer, but only at high cost—either inducing up to 77\% degenerate benign output or increasing over-refusal by up to 18\%. Because such degenerate benign outputs are not captured by standard refusal-based metrics, we introduce the \textbf{Benign Garble Rate} to quantify them. Our results suggest that cross-model robustness can be achieved by shaping representation geometry, without requiring attack-specific training. \smallskip \noindent\faGithub~Code, configs and logs: \href{https://anonymous.4open.science/r/AnchorRep/README.md}{anonymous.4open.science/r/AnchorRep}

Dynamic link prediction requires determining whether the historical interactions of a source node provide reliable structural support for a specific target. Existing target-conditioned methods capture such support through structural heuristics such as repeated interactions or co-neighbor overlap, but typically instantiate them as identity-based encodings and inject them as passive token-level features before aggregation. This design is brittle when exact structural encodings are sparse, and the resulting structural signals can be diluted after being concatenated with high-dimensional time and edge features, while their dynamics remain unmodeled. In this paper, we revisit target-conditioned dynamic link prediction from a structural dynamics modeling perspective and propose TCSD, which formulates structural signals as target-conditioned states, evolves them over recent histories, and uses them to guide aggregation. Concretely, we construct hybrid structural states from exact indicators that preserve time-aware repeat and co-neighbor matches, and relaxed indicators that recover latent support beyond exact identity matching. TCSD further models the dynamics of these states to capture temporal consistency within indicators and complementarity across indicators. Instead of treating structural states as passive features, TCSD uses the learned dynamics as aggregation controllers to amplify target-relevant interactions and suppress irrelevant ones. Experiments on sixteen dynamic graphs show that TCSD outperforms ten baselines, achieving up to a 14.27% relative improvement in MRR.

Online forecasters sometimes know a regime boundary before any post-boundary labels arrive. Standard adaptive conformal methods still react through later coverage errors, so they can pay a detection-delay cost even when the boundary is public. We study conformal prediction with an announced partition of the stream and separate two post-break audits. The reset empirical quantile in segmented split-CP attains the minimax per-time probability-gap rate $\Theta(M^{-1/2})$ after $M$ post-reset scores, with a matching Le Cam lower bound. A pure bounded constant-step announced-break ACI recursion attains an $O(M^{-1})$ time-averaged empirical-frequency gap by telescoping, but this is a realized-frequency statement, not a guarantee for the next interval. The practical AB-ACI implementation used in experiments adds burn-in and projection, so its rate claim includes explicit correction terms and is not unconditional. Synthetic experiments recover both slopes and show short-window gains when the break is large. Two real-data applications are null under leakage-free protocols. ERCOT RTC+B gives 0.863 post-reform coverage for AB-ACI versus 0.880 for ACI, and FOMC yield breaks give 0.892 versus 0.900. The message is diagnostic: use the reset empirical quantile for per-prediction reliability, use AB-ACI only for long-run frequency audits, and expect gains only when the realized score shift is large enough to make detection delay costly.

On-policy self-distillation, where a student is pulled toward a copy of itself conditioned on privileged context (e.g., a verified solution or feedback), offers a promising direction for advancing reasoning capability without a stronger external teacher. Yet in math reasoning the gains are inconsistent, even when the same approach succeeds elsewhere. A pointwise mutual information analysis traces the failure to the privileged context itself: it inflates the teacher's confidence on tokens already implied by the solution (structural connectives, verifiable claims) and deflates it on deliberation tokens (Wait, Let, Maybe) that drive multi-step search. We propose Anti-Self-Distillation (AntiSD), which ascends a divergence between student and teacher rather than descending it: this reverses the per-token sign and yields a naturally bounded advantage in one step. An entropy-triggered gate disables the term once the teacher entropy collapses, completing a drop-in replacement for default self-distillation. Across five models from 4B to 30B parameters on math reasoning benchmarks, AntiSD reaches the GRPO baseline's accuracy in 2 to 10× fewer training steps and improves final accuracy by up to 11.5 points. AntiSD opens a path to scalable self-improvement, where a language model bootstraps its own reasoning through its training signal.

Depth-of-field control is a fundamental tool in photography, yet post-capture bokeh editing from a single image remains challenging. A practical editor should handle images captured under arbitrary focus and aperture settings. Existing methods typically assume an all-in-focus input, or first recover an all-in-focus image before rendering new bokeh. Such pipelines can discard useful blur cues from the source image and propagate reconstruction artifacts into the final edit. We introduce AnyBokeh, a physics-guided framework for any-to-any bokeh editing. Instead of treating source blur merely as a degradation to be removed, AnyBokeh estimates the source blur state with a signed circle-of-confusion map and a disparity map. By modeling the linear relation between signed circle of confusion and disparity difference, AnyBokeh estimates a source-specific optical fingerprint and transfers the source optical characteristics to the desired focus and aperture setting. A generative editor conditioned on both source and target circle-of-confusion maps then performs relative blur synthesis, enabling spatially adaptive deblurring, preservation, and defocus rendering. To support physically supervised learning, we further construct a high-fidelity synthetic dataset with accurate depth, focus distance, and full EXIF metadata. Experiments on real-world benchmarks show that AnyBokeh achieves faithful and controllable editing across any-to-any bokeh editing, all-in-focus-to-bokeh rendering, and defocus deblurring, while avoiding all-in-focus reconstruction and test-time bokeh-level calibration commonly required by existing approaches. The code and dataset will be made publicly available.

Prediction sets provide a means of quantifying the uncertainty in predictive tasks. Using held out calibration data, conformal prediction and risk control can produce prediction sets that exhibit statistically valid error control in a computationally efficient manner. However, in the standard formulations, the error is only controlled on average over many possible calibration datasets of fixed size. In this paper, we extend the control to remain valid with high probability over a cumulatively growing calibration dataset at any time point. We derive such guarantees using quantile-based arguments and illustrate the applicability of the proposed framework to settings involving distribution shift. We further establish a matching lower bound and show that our guarantees are asymptotically tight. Finally, we demonstrate the practical performance of our methods through both simulations and real-world numerical examples.

Modern video generation models are increasingly hailed as emerging world models with an internalized grasp of physical law. Yet existing benchmarks largely evaluate physical plausibility only at the output level, without verifying whether the model arrives there through a faithful, law-grounded reasoning process. We introduce **Apple-$\pi$**, the first benchmark that anchors video-model evaluation explicitly in physical laws. Apple-$\pi$ comprises three components. **1) Orchard**: a dataset of 400 videos covering ten canonical tasks in classical mechanics. It separates single-law tasks for confounder-free diagnosis from multi-law tasks for probing generalization. **2) Benchmark Protocol**: a three-stage protocol based on scientific reasoning, including Perception, Formulation, and Deduction. It uses *chain-of-frames* prompting on infographic-annotated first frames, treating the generated video as the model's visible reasoning trace. **3) Evaluation Suite**: a hybrid evaluation suite that combines MLLM-based subjective scoring with physics-law-grounded objective measures. This enables stage-resolved diagnosis of not only whether a model fails, but where it fails. Benchmarking 11 models shows that current video models remain far from reliable law-grounded world simulators, with the best video model scoring only 0.473. Our stage-, pillar-, and source-resolved analyses further expose a Perception-to-Formulation-to-Deduction bottleneck, weak multi-law state transfer, and a persistent Sim-to-Real gap. These findings position Apple-$\pi$ as a diagnostic foundation for guiding future video models toward world models with law-grounded physical intelligence.

Contrastive representation learning (CRL) is a fundamental paradigm for learning transferable representations. Recent work identifies its population target as the pointwise mutual information (PMI) $ \log\frac{p(\mathbf{x},\mathbf{y})}{p_\mathcal{X}(\mathbf{x})p_\mathcal{Y}(\mathbf{y})} $ between two modalities $\mathcal{X},\mathcal{Y}$, where $p$ is the joint density function and $p_\mathcal{X},p_\mathcal{Y}$ are the marginals. However, it remains unclear why standard inner-product scores $f(\mathbf{x})^\top g(\mathbf{y})$ can effectively approximate a generally nonseparable population target PMI. In this paper, we address this question from the aspect of approximation for CRL. By linking PMI to a compact operator, we show that $D$-dimensional inner-product models perform rank-$D$ spectral approximation, with error controlled by the spectral tail. We then study neural network realizations: in general settings, we show that deep neural network scoring functions are universally consistent, and the approximation error rates depend on the smoothness of the PMI function; under structured low-rank data models we obtain improved rates attained by moderate-width encoders. Finally, we establish a fast calibration rate from contrastive excess risk to downstream retrieval error under a density-ratio Tsybakov noise condition.

Zeroth-order (ZO) optimization enables large-language-model fine-tuning without storing backpropagation activations, while LoRA supplies compact trainable adapters. Combining them creates a rank paradox: increasing LoRA rank improves adapter capacity, but standard two-point ZO either perturbs a rank-dependent number of coordinates or, under atomwise updates, can make the finite-difference signal unobservable. This paper shows that the bottleneck is a measurement-topology problem rather than a need for an external subspace. LoRA already decomposes into matched rank-$1$ atoms, each a complete factor-coordinate block of dimension $d_\text{out}+d_\text{in}$. Querying one atom per step keeps the stored adapter rank $r$ while removing $r$ from the single-query perturbation dimension. The naive atomwise query is still miscalibrated: if it inherits canonical LoRA scaling $\alpha/r$, the active finite-difference signal shrinks as $1/r$ and the active finite-difference signal-to-noise ratio (FD-SNR) as $1/r^2$, producing directional collapse under a fixed residual evaluation-noise floor. AR1-ZO pairs alternating rank-$1$ atom queries with topology-aware scaling $\gamma=\alpha r$, restoring rank-invariant active signal without auxiliary bases, activation hooks, curvature estimates, or extra forward queries. Theory proves atom minimality, rank-independent active query dimension, directional collapse and restoration, and the remaining rank dependence as an amortized coverage cost. Experiments on OPT and Qwen3 models validate the signal mechanism and show that AR1-ZO makes high-rank LoRA effective among matched-budget ZO methods under the standard two-forward-pass query budget.

No existing nested Archimedean copula tool handles all three of (a) arbitrary per-variable (right-)censoring in survival analysis, (b) arbitrary nesting trees, and (c) exact parameter gradients. Existing implementations handle only bivariate problems, low dimensional (i.e., $d \leq 10$) cases, two layers of nesting, or only hand-derived copula nestings. We present acopula, a JAX-native framework that, given any Archimedean generator—classical or neural—evaluates exact nested-copula likelihoods and parameter gradients under arbitrary censoring masks in polynomial time. The mechanism is polynomial powering of Taylor-mode automatic differentiation output, which replaces per-family hand-derived partial Bell polynomial tables with a single differentiable computation that any user-defined generator can drive. We conduct extensive simulations to verify the correctness of acopula. We then demonstrate (a) per-variable censoring on $85,229$ MIMIC-IV ICU admissions in high dimensions with $d=53$, fit by both classical Archimedean families and nested neural Archimedean copulas; (b) an 11-sector hierarchical model on S&P 500 daily returns at $d=98$; (c) family-agnostic censored MLE across ten families, five of them with no prior implementation, on a retinopathy study; and (d) a $\sim650\times$ per-density speedup over R's nacLL at $d=35$, scaling quadratically to $d=8,000$.

LLM-as-a-Judge pipelines have become the de facto evaluator for agent safety, yet existing benchmarks treat their verdicts as ground-truth proxies without checking whether the verdicts depend on the agent's behavior or merely on how the evaluation policy happens to be worded. We argue that any trustworthy safety judge must satisfy a basic property we call policy invariance, and we operationalize it as three testable principles: rubric-semantics invariance under certified-equivalent rewrites, rubric-threshold invariance under intentional strict-to-lenient shifts, and ambiguity-aware calibration so that verdict instability concentrates on genuinely ambiguous cases. Instantiating these principles as a stress-test protocol with four agent-class judges on trajectories drawn from ASSEBench and R-Judge, we surface a previously unmeasured failure mode: today's judges respond to meaningful normative shifts and to meaningless structural rewrites with comparable strength, and cannot tell the two apart. Content-preserving policy rewrites flip up to 9.1\% of verdicts above baseline jitter, and 18-43\% of all observed flips occur on unambiguous cases under such rewrites, so existing safety scores conflate what the agent did with how the evaluator was prompted. Beyond the diagnosis, we contribute the Policy Invariance Score and the Judge Card reporting protocol, which expose an order-of-magnitude spread in judge reliability that is invisible to accuracy-only leaderboards. We release the protocol and code so that future agent-safety benchmarks can audit their own evaluators rather than trust them by default: https://anonymous.4open.science/r/policy-invariance-judge

Text-to-image (T2I) generation models have achieved remarkable progress in producing visually realistic images from natural language prompts. Yet it remains unclear whether their success reflects genuine causal understanding or sophisticated pattern matching over visual-textual correlations. Inspired by Russell’s inductivist turkey, we introduce Counterfactual-World (CF-World), a counterfactual benchmark designed to probe whether text-to-image models can generate images under rules that systematically contradict real-world priors. CF-World organizes each scenario into three progressive levels: factual generation under ordinary world knowledge, explicit counterfactual generation with direct visual instructions, and implicit counterfactual generation requiring causal deduction from altered rules. We evaluate both open-source and closed-source T2I models using a Vision Language Model(VLM)-based evaluator (CF-Eval), and introduce two metrics: Prior Resistance Rate (PRR), which measures a model’s ability to overcome entrenched real-world priors, and Reasoning Retention Rate (RRR), which assesses whether models can maintain reasoning-dependent counterfactual generation without explicit visual cues. Experiments show that all models exhibit sharp degradation from factual to counterfactual settings. Further analyses suggest that these failures arise because current T2I models encode world knowledge and visual appearances as tightly coupled patterns. They rely heavily on frequent visual co-occurrences in their training data, causing them to default to familiar commonsense priors when tasked with generating counterfactual worlds.

Existing multimodal retrieval benchmarks largely emphasize semantic matching on daily-life images and offer limited diagnostics of professional knowledge and complex reasoning. To address this gap, we introduce ARK, a benchmark designed to analyze multimodal retrieval from two complementary perspectives: (i) knowledge domains (five domains with 17 subtypes), which characterize the content and expertise retrieval relies on, and (ii) reasoning skills (six categories), which characterize the type of inference over multimodal evidence required to identify the correct candidate. Specifically, ARK evaluates retrieval with both unimodal and multimodal queries and candidates, covering 16 heterogeneous visual data types. To avoid shortcut matching during evaluation, most queries are paired with targeted hard negatives that require multi-step reasoning. We evaluate 26 representative text-based and multimodal retrievers and observe a pronounced gap between knowledge- and reasoning-intensive retrieval, with fine-grained visual and spatial reasoning as persistent bottlenecks. We further show that enhancements such as re-ranking, rewriting, and agentic retrieval yield consistent gains, but substantial headroom remains. The dataset and evaluation pipeline will be released soon.

How much of zero-shot benchmark accuracy is captured by word-level alignment between a pre-training corpus and the benchmark target? We study this question through word-level unigram cross-entropy (UCE), a tokenizer-agnostic, model-free measure of lexical alignment between a reference corpus and a target benchmark. Likelihood training allocates probability mass according to the pre-training corpus, and likelihood-based zero-shot evaluation partly reads out the lexical alignment between this mass and the benchmark target. Cleanly tracing how the lexical imprint of pre-training data scales with corpus--benchmark alignment requires controlled from-scratch pre-training that varies corpus identity while holding the training recipe fixed. Across this controlled sweep, spanning $11$ zero-shot benchmarks, $4$ pre-training corpora, and $3$ model scales, we find that corpus--benchmark lexical alignment, UCE, consistently tracks zero-shot accuracy on every benchmark, with corpora more aligned with the benchmark (lower UCE) achieving higher accuracy. The same signal extends from measurement to control at finer scopes. At the document level, strengthening the imprint by selecting training documents to maximize lexical alignment with the benchmark yields a data-selection rule that improves accuracy over same-size random sampling and an importance-resampling baseline across sources and scales. At the prompt level, matching the imprint per prompt by routing each prompt to the drafter whose domain corpus is most lexically aligned with it gives a training-free routing rule for speculative decoding and improves throughput over a generalist-drafter baseline. Taken together, pre-training corpora leave a measurable lexical imprint on zero-shot behavior, and UCE provides a model-free probe of lexical alignment between reference and target data.

Recent years have witnessed tremendous progress in enabling LLMs to solve complex reasoning tasks such as math and coding. As we start to apply LLMs to harder tasks that they may not be able to solve in one shot, it is worth paying attention to their ability to construct intermediate stepping stones that prepare them to better solve the tasks. Examples of stepping stones include simplifications, alternative framings, or subproblems. We study properties and benefits of stepping stones in the context of modern reasoning LLMs via ARQ (Asking the Right Questions), a simple framework that introduces a question generator to the default reasoning pipeline. We first show that good stepping stone questions exist and are transferrable, meaning that good questions can be generated, and they substantially help LLMs of various capabilities in solving the target tasks. We next frame stepping stone generation as a post-training task and show that we can fine-tune LLMs to generate more useful stepping stones by SFT and RL on synthetic data.

Recent advances in machine learning and large-scale biological data collections have revived the prospect of building a virtual cell, a computational model of cellular behavior that could accelerate biological discovery. One of the most compelling promises of this vision is the ability to perform in silico phenotypic screens, in which a model predicts the effects of cellular perturbations in unseen biological contexts. This task combines heterogeneous textual inputs with diverse phenotypic outputs, making it particularly well-suited to LLMs and agentic systems. Yet, no standard benchmark currently exists for this task, as existing efforts focus on narrower molecular readouts that are only indirectly aligned with the phenotypic endpoints driving many real-world drug discovery workflows. In this work, we present AssayBench, a benchmark for phenotypic screen prediction, built from 1,920 publicly available CRISPR screens spanning five broad classes of cellular phenotypes. We formulate the screen prediction task as a gene rank prediction for each screen and introduce the adjusted nDCG, a continuous metric for comparing performance across heterogeneous assays. Our extensive evaluation shows that existing methods remain far from empirically estimated performance ceilings and zero-shot generalist LLMs outperform biology-specific LLMs and trainable baselines. Optimization techniques such as fine-tuning, ensembling, and prompt optimization can further improve LLM performance on this task. Overall, AssayBench offers a practical testbed for measuring progress toward in-silico phenotypic screening and, more broadly, virtual cell models. Our benchmark is available at https://github.com/Genentech/AssayBench.

Conditional generators provide a natural tool for controllable generation, including settings where the desired condition is a new composition of observed attributes or experimental factors. In many applications, especially in scientific domains, such models are attractive to explore conditions for which real samples are rare, expensive, or not yet observed. However, this creates a circularity for evaluation: standard conditional quality metrics require a reference target distribution, but in the extrapolative regime that distribution is unavailable by definition. We address this problem with a post-hoc, per-sample trust score for assessing conditional samples using only the training distribution. The score combines two estimable quantities: global realism, measuring compatibility with the real data manifold, and attribute-wise faithfulness, measuring whether a sample is closer to the requested attributes than to plausible alternatives. We show that the score can recover meaningful comparisons across extrapolated generations, under a mild coverage condition on the observed attributes. These comparisons enable effective filtering, ranking, and abstention of generations and can be used directly on off-the-shelf pretrained models. In biological imaging, selected samples preserve real morphological structure better and improve downstream predictive performance, while similar gains are observed on controlled vision benchmarks. Finally, we show how the score can be applied during generation, enabling abstention before full decoding.

Face Recognition systems are widely deployed in real-world applications, but they also raise privacy concerns due to unauthorized collection and misuse of facial data. Existing adversarial privacy protection methods rely on input-space perturbations to obfuscate identity information, yet their protection can degrade when adversaries learn restoration or purification mappings that partially invert the transformation. We study this setting as an asymmetric adversarial attack, in which reverse manipulation becomes feasible because existing defense paradigms do not control reversibility. To address this problem, we propose Asymmetric Reversible Face Protection (ARFP), a restoration-aware extension of personalized face cloaking that integrates privacy protection, keyed recovery, and tamper indication in a single framework. ARFP consists of three components: Key-Conditioned Manifold Binding, which ties the protection transformation to a user-provided key; Adversarial Restoration-Aware Training, which introduces a surrogate restoration adversary during training to improve robustness against evaluated inverse purification attacks; and Authorized Reversible Restoration, which supports recovery with the correct key while providing nonce-based tamper indication. Extensive experiments under the threat models considered in this work show that ARFP improves resistance to the evaluated restoration attacks while preserving authorized recovery utility. These results provide empirical evidence of key-sensitive recovery behavior and tamper awareness in the tested settings.

LLM agents are increasingly studied as systems that can self-evolve during deployment. A representative mechanism for realizing self-evolution is memory evolution, where agents maintain evolving memories derived from previous tasks and condition future inference on their memory. Although memory evolution is intended to improve agent performance, it also creates a feedback loop in which model-generated content can persist across tasks and influence future decisions. We study this loop as a delayed attack channel for memory-evolving agents. We introduce asynchronous agentic poisoning, a provider-side threat in which a malicious provider releases a model that appears safe under fresh-context evaluation but exhibits unsafe behavior after self-evolving through memory evolution. We realize this threat by fine-tuning the released model to silently embed a hidden marker into its otherwise normal responses, causing the marker to accumulate in the agent’s memory through memory evolution. When the accumulated markers are later retrieved into the input context, the model reacts to them as a backdoor trigger and shifts to attacker-specified behavior. This creates a self-triggered transition from safety-aligned behavior to attacker-specified behavior without requiring any post-deployment attacker interaction. Experimental results show that our method increases the harmful rate from below 5\% to around 90\% after memory evolution, while not substantially degrading the model’s self-evolution performance. These findings demonstrate that memory evolution creates a provider-side pathway for delayed compromise, allowing a model to appear safe at release time while becoming unsafe after deployment through the agent’s self-evolution process.

We study language generation in the limit under a global preference ordering on strings, as introduced by Kleinberg and Wei. We aim for breadth, but impose an additional requirement of timeliness: higher-ranked strings should be generated earlier. A string is then only credited if it is generated before a deadline, where its deadline is defined by a function that maps a string’s rank in the target language to the time by which it must be produced. This is in keeping with a central consideration in machine learning, where inductive bias favors "simpler" or "more plausible" outputs, all else being equal. We show that timely generation is impossible in a strong sense for eventually consistent generators—the protagonists of most prior related work. Under what is perhaps the mildest natural relaxation of consistency, a hallucination rate that vanishes over time, we show that we can circumvent our impossibility result. In particular, we can achieve optimal density with respect to any superlinear deadline function. We also show this is tight by ruling out timely generation with linear deadlines and vanishing hallucination rate.

Understanding the dynamic behavior of biomolecules is fundamental to elucidating biological function and facilitating drug discovery. While Molecular Dynamics (MD) simulations provide a rigorous physical basis for studying these dynamics, they remain computationally expensive for long timescales. Recent deep generative models accelerate conformation generation but often either discard temporal correlations entirely or struggle to condition on the extended history required for faithful kinetics, a consequence of the effectively non-Markovian nature of partially observed subsystem coordinates. To bridge this gap, we introduce ATMOS, a novel generative framework based on State Space Models (SSM) designed to generate atom-level MD trajectories for biomolecular systems. ATMOS integrates a Pairformer-based state transition mechanism to capture temporal dependencies, with a diffusion-based module to decode trajectory frames autoregressively. We demonstrate that ATMOS achieves state-of-the-art performance in generating conformation trajectories for both protein monomers and protein-ligand systems. This work provides a unified and computationally efficient framework for biomolecular trajectory generation, taking a step toward dynamics foundation models.

Graph-structured optimization with linear constraints is fundamental to critical infrastructure but faces scalability limits due to massive strict hard constraints and high dimensionality. While recent projection-based methods such as Trainable Sampling Kaczmarz-Motzkin Net (T-SKM-Net) guarantee feasibility, they face high computational costs in dynamic environments by processing the entire constraint set and requiring expensive matrix factorizations. To bridge this gap, we propose the Accelerated Trainable-SKM (AT-SKM) Net framework. To concentrate computation on the active constraints and eliminate redundant calculations, we introduce a hybrid sampling strategy guided by a topology-aware heterogeneous GNN model. To efficiently handle topological shifts in graph-based constraints, we employ a Cholesky Update mechanism that theoretically reduces the equality projection complexity from $\mathcal{O}(N^3)$ to $\mathcal{O}(N^2)$ under low-rank perturbations. Experiments on random geometric graphs, N-1 Security-Constrained DC-OPF, and minimum-cost gas transport problem demonstrate that AT-SKM reduces iteration counts by up to 85% and achieves 2.95$\times$-7.29$\times$ SKM layer speedups, while maintaining zero constraint violations. Our code is available at [https://anonymous.4open.science/status/Anonymous_Submission-CFC7](https://anonymous.4open.science/status/Anonymous_Submission-CFC7)

Modern video object segmentation (VOS) relies on a two-stage paradigm for robust performance: target initialization and memory-based propagation. However, existing adversarial attacks on VOS often overlook this fundamental feature. To bridge this gap, we propose AUV, an adversarial framework that Aligns with the Universal two-stage paradigm of VOS models. Specifically, AUV optimizes a single universal adversarial perturbation (UAP) through decoupled training strategies tailored for both initialization and propagation stages. To ensure prompt-agnosticism across all modalities, including masks, points, boxes, and language, AUV optimizes the UAP to disrupt the shared memory space for latent-level target erosion. Extensive experiments show that AUV establishes a new state-of-the-art for VOS attacks, such as degrading mask-based SAM3 to only 8.4 J&F on DAVIS17-val (compared to 55.4 for previous SOTA under the same training setup), while consistently outperforming prior methods on other VOS models including SAM2, Cutie, and XMem. Notably, AUV exhibits superior temporal dynamics with the fastest onset and the most enduring persistence, while uniquely supporting flexible any-time mid-video attack, revealing vulnerabilities in real-world streaming applications. Our code and checkpoint will be made publicly available.

Visual perception depends on top-down goals and bottom-up sensory mechanisms. Vision-language models implement both, allowing us to treat each component as a separable hypothesis about what drives where we look. We compared spatial attention maps from six vision-language models against human fixation heatmaps recorded on 200 images during two tasks (general description and social captioning). The six models spanned a 2$\times$2 factorial of CNN vs.\ ViT encoders crossed with LSTM vs.\ Transformer decoders, plus Molmo 7B-D and Qwen3.5 9B. We found that both decoder and encoder architecture shaped alignment, but decoder choice dominated. LSTM vs.\ Transformer decoders increased alignment by 40--50 percentage points (80--87\% vs.\ 40--59\% of the human noise ceiling). In contrast, CNN vs.\ ViT encoders contributed a secondary 5--20 point advantage depending on decoder family, with CNN-LSTM the most aligned model overall (85--87\%). Despite their alignment advantage, LSTM-decoder attention maps were spatially diffuse and minimally task-differentiated; ViT-Transformer, the weakest in alignment, showed the sharpest spatial concentration and strongest task differentiation. A hemispatial-neglect simulation confirmed that ablating attention impacted LSTM decoders more than Transformer decoders. In an exploratory extension using TRIBE-simulated synthetic neural responses, fixation alignment and neural relevance dissociate: CNN-Transformer attention maps better predicted synthetic brain activity despite lower fixation alignment, with attention maps best predicting early visual cortex. Together, top-down and bottom-up components trade off what they predict in behavioral and synthetic neural data.

Auto-regressive models (ARMs) have established a dominant paradigm in language modeling. However, their strictly sequential sampling paradigm imposes fundamental constraints on both inference efficiency and modeling flexibility. To address these limitations, diffusion-based large language models (dLLMs) have been proposed, offering the potential for parallel sampling and flexible language modeling. Despite these advantages, current dLLMs sampling strategies rely primarily on token level information, which fails to account for global sequence structure and often yields suboptimal results. In this paper, we study the sampling order selection problem from the perspective of log-likelihood maximization. We show that this problem is NP-hard and propose an optimal sampling-rank-based approximation that makes the objective computationally tractable. We further prove that the tractable objective is optimized by sampling tokens in descending order of their attention-matrix column sums. This finding provides a principled justification for attention-guided sampling and offers a theoretically grounded alternative to greedy search. We instantiate this theoretical insight in a new training-free sampling algorithm, termed Attn-Sampler, and further propose dynamic attention thresholding for practical acceleration. Extensive experiments across multiple benchmarks validate the effectiveness of our proposed method, demonstrating that it achieves superior generation quality while enhancing the sampling parallelism.

Attention sinks and massive activations are recurring and closely related phenomena in Transformer models. Existing explanations have largely focused on the forward pass, yet in pre-norm Transformers, large residual-stream norms play only an indirect forward role because sublayers operate on normalized inputs. We study this relationship from the perspective of backpropagation. Empirically and theoretically, we show that under causal masking, attention sinks can induce pronounced gradient concentration, which we term gradient sinks. Since the RMSNorm Jacobian attenuates gradients roughly in inverse proportion to input norm, massive activations can be understood as adaptive regulators of this localized gradient pressure during training. This interpretation predicts that attenuating sink-induced gradients should weaken massive activations. We test this prediction with V-scale, a modification that adjusts backpropagated gradients on the value path. In V-scale models, attention sinks are preserved, whereas massive activations are suppressed. These results identify gradient sinks as a backward-pass counterpart of attention sinks, and massive activations as an adaptive RMSNorm-mediated response that attenuates the resulting localized training pressure. Our code is available at https://anonymous.4open.science/r/GradientSinkCode-B309.

Cross-architecture/backbone medical-segmentation pools are commonly motivated as complementary failure coverage when members make different mistakes; that assumption is rarely tested at the per-case level. We audit this assumption for a fixed pool of 11 pretrained-encoder bundles read out by a deliberately weak $1\times1$-convolution decoder (4 seeds), and for a small set of decoder-recipe perturbations on the same bundles. Across five primary task rows (Kvasir polyp, ACDC LV, BraTS foreground, RIGA Cup, RIGA Disc), this 11-bundle $1\times1$-conv readout pool yields per-case Dice correlation gaps of $\Delta=0.261\text{--}0.638$ between same-bundle and cross-bundle pairs after a Dice $\geq 0.30$ functional floor; all paired case and hierarchical bootstrap intervals exclude zero. The result is stable under floor sweeps, ICC(A,1), leave-one-out, subject aggregation, quality-controlled pair regression, and cross-fitted item-difficulty residualisation. Decoder-recipe diagnostics show that richer readouts attenuate the gap (UNet-skip; case-identical nnU-Net on RIGA Cup: $0.473\to 0.0079$, separate-split $0.034\text{--}0.047$); the reported magnitudes are recipe-specific. Additional probes (pixel error, joint-failure lift, same-architecture-family ViT, same-family cross-checkpoint, same-backbone) locate the dependence structure but do not causally separate architecture, pretraining, and recipe. The claim is scoped to 2D frozen/light-adaptation encoder pools with this readout family and to retrospective benchmark audits. We release per-case Dice traces, summary JSONs, scorer code, hashes, and metadata as the audit artifact.

Tabular Foundation Models (TabFMs) produce context-aware embeddings that are increasingly stored, shared, and reused for retrieval, clustering, and cross-institutional data exchange. These embeddings are often treated as privacy-preserving substitutes for raw tabular records, yet their attribute-level information content remains poorly understood. We present a systematic privacy audit of TabFM embeddings: given an embedding and varying amounts of side information, how accurately can sensitive attributes be recovered? We introduce Cascade Probing, a sequential probing method that measures recoverability while accounting for inter-attribute dependencies, significantly outperforming joint probing and optimization-based inversion baselines. Across four TabFMs (TabPFN, TabDPT, Mitra, TabICL) and eight datasets, we find that even without any side information, 65-95\% of sensitive attributes can be recovered from embeddings alone. More critically, recoverability exhibits a Privacy Cliff: revealing a single known attribute sharply increases the recoverability of all remaining attributes. This phenomenon generalizes across model architectures and diverse domains, and is not eliminated by standard embedding transformations including noise injection, PCA, and knowledge distillation. These findings suggest that we need more formal privacy techniques for protecting sensitive information inside TabFM embeddings.

While Large Language Models (LLMs) are increasingly deployed as automated evaluators in evaluation and training pipelines, their judgements are often affected by systematic biases that conflict with human preferences. While prior work has identified several known biases and proposed methods for their detection and mitigation, they lack strong grounding in human evaluation preferences, which is essential to ensuring that the identified biases correspond to actual human judgment behavior. Moreover, they rely heavily on pre-discovered bias lists, overlook bias strength, and depend on costly interventions. In this work, we propose HUB-J, an integrated framework grounded in human evaluation preferences that detects, quantifies, and mitigates biases in LLM-as-a-Judge systems. Our approach leverages human–LLM judgement disagreement cases to automatically discover interpretable bias factors, and utilizes agreement cases to quantify bias strength through controlled input modifications and resulting shifts in model decisions. Finally, building on these quantified biases, we introduce a lightweight, training-free regression-based mitigation strategy that corrects bias-influenced judgments by removing the estimated bias effects. Empirical results show that HUB-J uncovers both known and novel bias factors, reveals meaningful differences in model susceptibility, and consistently reduces bias-driven decision flips while generalizing across models.

Predictive coding (PC) is a local-learning alternative to backpropagation (BP), training deep networks via a local energy-minimization dynamics rather than a global backward pass. We introduce Augmented-Lagrangian Predictive Coding (PC-ALM), which maintains PC's inference budget but aligns each weight update toward BP by accumulating per-layer constraint errors into a layer-local Lagrange multiplier. In linear PC networks, PC-ALM converges to an equilibrium with exact BP gradients distributed across the network via only layer-local updates. We analyze PC-ALM in nonlinear PC networks up to depth 128 and show that it matches BP performance across all width-depth regimes, notably in deep narrow networks where PC underperforms. PC-ALM introduces recurrent dynamics in each layer's activations. Compared to PC's heat flow on a scalar energy, PC-ALM dynamics are driven by dual ascent on the Augmented Lagrangian. We observe ballistic credit propagation across especially deep networks, with prediction errors evenly distributed across layers, compared to PC's diffusive credit propagation. Beyond the algorithm itself, the augmented Lagrangian framework offers a generalization of PC, and may yield insights on how distributed systems could compute and propagate BP-like credit signals through purely local dynamics.

Differential Privacy (DP) bounds the privacy leakage of a mechanism against worst-case membership inference, but the precise tradeoff between complex adversarial models and DP protections remains poorly understood. In this paper, we present a unified framework that generalizes the patchwork of existing bounds across membership inference, attribute inference, and data reconstruction attacks. Crucially, our framework is the first to evaluate attacks that target multiple individuals simultaneously and measure success beyond exact matches under a single cohesive bound. Our bounds capture this broad family of previously unexplored attack settings by relying solely on the privacy parameters and the adversary's baseline success rate (i.e. its prior without access to the mechanism's output). To illustrate this, we compare our high-probability guarantees to empirical attacks in two novel settings: extracting multiple non-uniform secrets (passwords and PII) from DP-finetuned language models, and reconstructing tabular data from noisy marginals. Ultimately, this framework provides a rigorous theoretical foundation to investigate the risk landscape of DP algorithms in new adversarial settings.

A wide range of methods have been proposed for interpreting language models, delivering important insights into their inner workings. However, different methods and their resulting insights stand in relative isolation: what could the underlying structure of language models be, such that they give rise to all our interpretations? In this work, we propose using Tensor Product Representations (TPRs; Smolensky 1990) as a unifying hypothesis. TPRs give a concrete proposal for how compositional structure can be represented in vector space. We show, both mathematically and empirically, that TPRs can unify several prior interpretability methods: additive analogies, linear probing, sparse autoencoders, and activation patching. Mathematically, we show that these methods can all be derived from TPRs. Empirically, we apply the derivations to a range of different models — from small toy models to LLMs — to construct instances of each of the above interpretability methods; these constructed variants perform comparably to their standard variants. We view this work as a step toward what interpretability will ideally provide: a unified account of the nature of neural networks, corroborated not just by individual observations but also by an explanation of the connections between them.

Real-world digital environments are highly diverse and dynamic. These characteristics cause agents to frequently encounter unseen environments and distribution shifts, making continual learning in such environments essential for computer-use agents (CUAs). However, a key challenge lies in obtaining high-quality and environment-grounded training data without relying on costly human annotation. In this work, we introduce ACuRL, an Autonomous Curriculum Reinforcement Learning framework that continually adapts agents to specific environments with zero human data. The agent first explores an environment to acquire initial experiences. During subsequent iterative training, a curriculum task generator leverages these experiences together with feedback from the previous iteration to synthesize new tasks tailored for the agent's current capabilities. To provide reliable reward signals, we introduce CUAJudge, a robust automatic evaluator for CUAs that achieves 93% agreement with human judgments. Empirically, our method effectively enables both intra-environment and cross-environment continual learning, yielding 3–29% absolute performance gains on the target environments without catastrophic forgetting on others. We also show that it can mitigate performance degradation under environment changes (e.g., version updates, platform migration, and resolution shifts). Further analyses show highly sparse updates (e.g., only 20% parameters), which helps explain the effective and robust adaptation.

Visual mental imagery (VMI) decoding uses brain activity to reconstruct internally generated visual scenes, providing a potential pathway for externalizing internal experiences and developing neuroprosthetic communication interfaces. However, progress in this field is limited by the lack of dedicated, training-scale VMI datasets. Current VMI paradigms often rely on visual prompts or short-term memory images, introducing visual and mnemonic confounds that make it difficult to disentangle neural activity associated with endogenous image construction from signals driven by prior or concurrent visual input. Furthermore, existing VMI datasets offer limited descriptive control over individual visual attributes, making it difficult to systematically assess fine grained visual details. To address these gaps, we introduce AVID, the first dedicated training-scale 5.0 Tesla functional magnetic resonance imaging (fMRI) resource for auditory-induced VMI decoding. We further define a benchmark around AVID with fixed splits, predefined neural inputs, and a standardized evaluation protocol. Acquired with high-field 5T fMRI, AVID comprises approximately 67 hours of dense recordings from nine participants. The dataset is uniquely structured into two complementary splits: a scene-level complex split using naturalistic Mandarin descriptions and a controlled simple split with explicitly specified visual attributes. We further formalize a standardized evaluation protocol that decouples source-image consistency from semantic alignment, allowing for a more nuanced assessment of imagery reconstruction. Baseline qualitative and quantitative results show that current reconstruction pipelines remain limited on AVID, although the most successful reconstructions preserve recognizable scene level semantics from spoken descriptions. By combining dedicated training-scale neural recordings and standardized evaluation resources, AVID provides both a data foundation for imagery-specific model development and a benchmark for evaluating VMI decoding with substantially reduced sensory contamination.

Multimodal Large Language Models have achieved remarkable progress in short-form audio-video understanding, yet long-form audio-video comprehension remains challenged by limited context windows and severe information redundancy. To address these bottlenecks, we propose AVOC, a framework for long-form audio-video understanding in Omni-modal Large Language Models. AVOC introduces a learnable token compression module between the modality encoders and the LLM backbone. We reframe multimodal token compression as a top-$K$ retrieval problem: given a fixed context budget, the module must retrieve a compact subset of tokens that best supports answering the user query. We draw inspiration from three classical Information Retrieval criteria for selecting informative units from a large candidate pool: relevance, importance, and diversity. AVOC instantiates each criterion as a tailored mechanism for audio-video understanding, and integrates them into a unified retrieval-style compression pipeline. Experiments show that AVOC achieves state-of-the-art performance on long-form audio-video benchmarks, surpassing the second-best model by 4.9 and 5.5 points in average accuracy on OmniVideoBench and LVOmniBench, respectively. Moreover, AVOC maintains robust performance on Audio-Video Needle-in-a-Haystack task at durations up to one hour.

Training powerful AI systems to exhibit desired behaviors hinges on the ability to provide accurate human supervision on increasingly complex tasks. A promising approach to this problem is to amplify human judgement by leveraging the power of two competing AIs in a debate about the correct solution to a given problem. Prior theoretical work has provided a complexity-theoretic formalization of AI debate, and posed the problem of designing protocols for AI debate that guarantee the correctness of human judgements for as complex a class of problems as possible. Recursive debates, in which debaters decompose a complex problem into simpler subproblems, hold promise for growing the class of problems that can be accurately judged in a debate. However, existing protocols for recursive debate run into the obfuscated arguments problem: a dishonest debater can use a computationally efficient strategy that forces an honest opponent to solve a computationally intractable problem to win. We mitigate this problem with a new recursive debate protocol that, under certain stability assumptions, ensures that an honest debater can win with a strategy requiring computational efficiency comparable to their opponent.

Multi-task learning requires a shared update that makes balanced progress across multiple objectives. A key obstacle is that progress is difficult to compare across tasks with different loss scales, units, and local optimization dynamics. We study this problem from an optimality-gap perspective. For each task, we compare the improvement induced by the shared update with the best one-step improvement achievable by optimizing that task in isolation. This comparison yields a scale-invariant normalized improvement rate, which measures how well the shared update serves each task relative to its own local improvement bound. We formulate balanced multi-task optimization as a max-min problem over the normalized improvement rates, thereby prioritizing the worst-served task. A local quadratic approximation leads to the Karush-Kuhn-Tucker (KKT) optimality conditions, showing that the optimal shared update is determined by a small set of bottleneck tasks. Under an isotropic Hessian approximation, the update admits a closed-form expression that depends only on task gradients and a regularized Gram matrix, while the bottleneck set is identified by an active-set procedure. Experiments on diverse benchmarks show improved task balance and competitive performance.

Existing AI benchmarks lack the fidelity to assess economically meaningful progress on professional workflows. To evaluate frontier AI agents in a high-value, labor-intensive profession, we introduce BankerToolBench (BTB): an open-source benchmark of end-to-end analytical workflows routinely performed by junior investment bankers. To develop an ecologically valid benchmark grounded in representative work environments, we collaborated with 502 investment bankers from leading firms. BTB requires agents to execute senior banker requests by navigating data rooms, using industry tools (market data platform, SEC filings database), and generating multi-file deliverables—including Excel financial models, PowerPoint pitch decks, and PDF/Word reports. Completing a BTB task takes bankers up to 21 hours, underscoring the economic stakes of successfully delegating this work to AI. BTB enables automated evaluation of any LLM or agent, scoring deliverables against 100+ rubric criteria defined by veteran investment bankers to capture stakeholder utility. Testing 9 frontier models, we find that even the best-performing model (GPT-5.4) fails nearly half of the rubric criteria and bankers rate 0% of its outputs as client-ready. Our failure analysis reveals key obstacles (such as breakdowns in cross-artifact consistency) and improvement directions for agentic AI in high-stakes professional workflows.

We study Bayesian optimization (BO) through the lens of information geometry. Pulling back the Fisher information metric through the surrogate posterior map yields a local sensitivity tensor on the input space, which leads to an upper bound on the gradient of reparameterizable acquisition functions. This view explains vanishing-gradient behavior in high-dimensional BO and provides a common interpretation of heuristics such as RAASP and dimension-scaled lengthscales. Building on this analysis, we propose FITR, a trust-region-based BO method that replaces lengthscale-based scaling by local pullback-Fisher weights. FITR is not restricted to GP kernels with explicit lengthscales. On GP benchmarks with an SE kernel, experiments show competitive performance using FITR. The proposed method also easily generalizes to non-isotropic surrogates, although the gains are more task-dependent in that setting.

Multi-view video recordings are increasingly used to capture the 3D movements of animals in experimental settings, yet extracting rich 3D representations from these recordings remains challenging. Supervised pose estimation requires extensive manual annotation, while general-purpose 3D reconstruction models trained on generic scene datasets fail on the specialized imagery and sparse-view setting of laboratory experiments. We address these limitations with BEAST3D, a self-supervised pretraining framework that learns 3D visual representations from unlabeled, calibrated multi-view video. BEAST3D uses a vision transformer to predict 3D Gaussian splats that reconstruct held-out views through differentiable rendering, while simultaneously segmenting the animal from the background. BEAST3D reconstructs 3D structure with as few as four views by conditioning directly on known camera parameters---unlike general-purpose models, which must estimate camera geometry from dense overlapping viewpoints that are seldom available in lab settings. Through comprehensive evaluation across four species, we demonstrate that BEAST3D produces rich, viewpoint-invariant features that transfer effectively to three downstream tasks: novel view synthesis, which validates the quality of the learned 3D representations; multi-view pose estimation, which provides the sparse keypoint trajectories widely used in behavioral analysis; and neural encoding, which relates 3D behavioral features to simultaneously recorded neural activity. BEAST3D thus establishes a versatile framework for behavioral analysis that leverages 3D structure in modern multi-view laboratory recordings.

Children's earliest years are a period of intense social learning, during which meaning is built through gaze, gesture, and physical contact. Yet, this critical aspect of social interaction remains under-evaluated in multimodal video understanding, as existing benchmarks prioritize the conversation-driven interactions of adults. We address this gap with NEST ($\textbf{N}$aturalistic $\textbf{E}$arly-childhood $\textbf{S}$ocial in$\textbf{T}$eractions), the first child-centric benchmark for nonverbal social reasoning. NEST comprises 1,208 manually annotated clips ($\sim$12s) curated for diversity across naturalistic settings, cultural contexts, and developmental stages. Beyond a lack of data, targeted model improvement in this domain is hindered by coarse evaluation schemas that conflate basic perception with higher-order reasoning. To resolve this, NEST introduces a compositional schema that builds atomic interaction cues into progressive levels of social reasoning: contextual, behavioral, interpersonal, and open-ended descriptions. Extensive analysis of state-of-the-art Vision-Language Models (VLMs) reveals a stark performance collapse: despite strong contextual recognition, models exhibit substantial gaps in decoding precise behavioral and interpersonal cues. We trace these failures to specific biases, including poor consequence grounding and temporal sensitivity. Finally, we ground NEST in real-world applications by evaluating VLMs using domain adaptation strategies. Ultimately, NEST provides a rigorous testbed for advancing socially grounded artificial intelligence in early childhood. The dataset and code will be publicly released.

Markov decision problems are most commonly solved via dynamic programming. Another approach is Bellman residual minimization, which directly minimizes the squared Bellman residual objective function. However, compared to dynamic programming, this approach has received relatively less attention, mainly because it is often less efficient in practice and can be more difficult to extend to model-free settings such as reinforcement learning. Nonetheless, Bellman residual minimization has several advantages that make it worth investigating, such as more stable convergence with function approximation for value functions. While Bellman residual methods for policy evaluation have been widely studied, methods for policy optimization (control tasks) have been scarcely explored. In this paper, we establish foundational results for the control Bellman residual minimization for policy optimization.

LLM judges are increasingly placed inside an agent's loop, scoring the agent's own attempts and re-prompting until one passes. We show this quietly corrupts measurement: retry-until-PASS is optional stopping against a noisy classifier—it keeps drawing until the judge slips—so the reported pass rate is a biased estimator of true success, upward in the pass-prone retry regimes of interest (and downward under conservative rules such as strict rubrics or unanimous juries). We make this exact. The cap-K gate is a binary classifier with closed-form sensitivity/specificity, and its bias is governed by one coefficient, the gate's Youden index J: as J → 0 the gated rate becomes uninformative about π, so recovery must fall back to gold labels and no estimator beats the gold-only mean. Across 44 capable-agent loops on GSM8K, MATH, and code with objective ground truth (no authored weakness; a separate terse-agent stress set is excluded here) the inflation is systematic (median slip +0.16; worst on code, where the judge cannot run the candidate, a true 0.74 inflated to a reported 0.98) and obeys a closed-form law predicting the slip from per-attempt statistics (pooled r = 0.95; errors-in-variables slope 0.765 [0.70, 0.84], excluding 1). To recover true success we benchmark Rogan–Gladen against prediction-powered inference: the label-efficient estimators (PPI/PPI++) recover π far more accurately than naive reporting and Rogan–Gladen (mean recovery MAE 0.050 vs. 0.149, and 0.081 vs. 0.241 as gold becomes scarce), because they escape the 1/J² variance that makes the classical correction fragile; on balanced low-bias gates they match the gold-only mean (per-gate differences within bootstrap noise, none uniformly best), so our contribution is not a uniquely-best estimator but the identification of loop-induced bias and the J diagnostic that says when—and whether—to correct at all. Beyond verifiable gold, we measure recovery on public, human-labeled non-verifiable gates—response safety and summary quality—where PPI++ recovers the true rate to mean MAE 0.043 versus naive 0.165 (~4× in aggregate, up to >10× on the most-biased gates); and—in the motivating regime, a non-verifiable safety judge inside a real retry loop (n = 400, against a pre-registered 3-model strong-LLM panel—a disclosed proxy, human-anchored at raw 0.90 agreement)—a lenient gate ships 6.8% panel-labeled-unsafe responses (95% CI [0.047, 0.096]) while the calibrated correction recovers the panel safe-rate ~3.5× more accurately than naive. The deliverable is a recipe: report a PPI++ estimate alongside J as a reliability/identifiability diagnostic, measure at K = 1, and—via a label-free drift detector (ROC-AUC 0.80)—de-bias only when calibration transfers. We release all code and content-free data.

Here we introduce the Olfactory Perception (OP) benchmark, designed to assess the capability of large language models (LLMs) to reason about smell. The benchmark contains 1,010 questions across eight task categories spanning odor classification, odor primary descriptor identification, intensity and pleasantness judgments, multi-descriptor prediction, mixture similarity, olfactory receptor activation, and smell identification from real-world odor sources. Each question is presented in two prompt formats, compound names and isomeric SMILES, to evaluate the effect of molecular representations. Evaluating 21 model configurations across major model families, we find that compound-name prompts consistently outperform isomeric SMILES, with gains ranging from +3.1 to +18.9 percentage points (mean $\approx$ +7 points), suggesting current LLMs access olfactory knowledge primarily through lexical associations rather than structural molecular reasoning. The best-performing model reaches 64.4\% overall accuracy, which highlights both emerging capabilities and substantial remaining gaps in olfactory reasoning. We further evaluate a subset of the OP across 21 languages and find that aggregating predictions across languages improves olfactory prediction, with AUROC=0.86 for the best performing language ensemble model. LLMs should be able to handle olfactory and not just visual or auditory information.

The recent development of Large Language Models (LLMs) has been accompanied by an effervescence of novel ideas and methods to better optimize the loss of deep learning models. Claims from those methods are myriad: from faster convergence to removing reliance on certain hyperparameters. However, the diverse experimental protocols used to validate these claims make direct comparisons between methods challenging. This study presents a comprehensive evaluation of recent optimization techniques across standardized LLM pretraining scenarios, systematically varying model size, batch size, and training duration. Through careful tuning of each method, we provide guidance to practitioners on which optimizer is best suited for each scenario. For researchers, our work highlights promising directions for future optimization research. Finally, by releasing our code and making all experiments fully reproducible, we hope our efforts can help the development and rigorous benchmarking of future methods.

Biosynthetic gene clusters (BGCs) are co-located genes that encode the biosynthesis of secondary metabolites, a major source of clinically used antibiotics and anticancer agents. Recent advances in deep learning, particularly self-supervised foundation models, have spurred growing interest in BGC sequence modeling, but evaluation infrastructure has lagged behind. Current practice often relies on coarse classification accuracy over small, experimentally curated datasets, making it difficult to discriminate model capabilities or assess downstream utility. We introduce BGC-Bench, an evaluation suite designed to resolve this limitation by systematically incorporating biochemical knowledge in the benchmark design. In terms of generalization, we test whether representations transfer to genetically and chemically novel samples; for functionality, we examine application-relevant tasks, including retrieval, prediction, and active learning. BGC-Bench reveals that no single model class dominates: pretrained foundation models perform strongly under out-of-domain generalization but remain sensitive to training conditions; compositional baselines are competitive and lead in cross-modal retrieval; and active-learning gains vary substantially across tasks, strategies, and model families. Beyond these BGC-specific findings, BGC-Bench highlights a broader principle: domain-informed benchmark design over existing datasets enables better evaluation in data-scarce scientific domains.

Diffusion models achieve strong generative performance but often struggle to align generated samples with human preferences measured by a reward model. A simple yet effective algorithm for test-time alignment is Best-of-$N$ (BoN) sampling, which draws $N$ i.i.d. samples from a pre-trained diffusion model and outputs the single highest-reward sample. Despite its empirical success, this procedure is inherently post hoc: reward information is used only for selection after generation, not for improving the reverse diffusion process itself. Consequently, BoN sampling does not improve the average alignment of generated samples and is primarily suited to single-output settings. We propose Best-of-$N$ Guidance (BoNG), a novel method that integrates the principle of BoN sampling directly into the reverse diffusion process. BoNG performs online BoN selection over denoising particles and adjusts the reverse diffusion process to steer the particle population toward higher-reward regions during generation. Specifically, by introducing an asymmetric guidance interaction among denoising particles, BoNG uses the current BoN particle as a guidance signal to the rest of the particle population. This particle-level interaction reshapes the sampling process toward higher-reward regions, enabling BoNG to improve not only the final best sample beyond Vanilla BoN sampling, but also the average quality of generated samples. Over 36 empirical comparisons, BoNG achieves the best performance in 29 cases, ranking first in 80.56\% of the comparisons against SMC and Vanilla BoN sampling. BoNG also supports multi-output capability, showing 1.3$\times$ ImageReward score than the latest sample-based guidance method with 1.6$\times$ speedup.

The Message Passing Neural Networks (MPNNs) and Graph Transformers (GTs) have emerged as two dominant paradigms for graph representation learning. However, the expressive power of standard MPNNs is fundamentally bounded by the 1-dimensional Weisfeiler-Lehman (1-WL) test, while GTs lack the inductive bias for graph structure. To enhance structural representation, existing methods typically resort to subgraph-based aggregation or coordinate-based positional encodings. However, the former suffers from prohibitive computational memory overheads, while the latter is limited by rigid reference frames that fail to distinguish fine-grained local topology. To address these limitations, we introduce a novel structural encoding framework based on contextual distributions. Specifically, we move beyond fixed coordinate systems to capture intrinsic local topology by encoding structural information through a compact distributional statistic, i.e., the entropy of the node context. Furthermore, instead of relying on relative positioning, we introduce a kernelized mechanism to encode relational similarity by quantifying the structural affinity between node contexts. Extensive experiments on both synthetic and real-world datasets demonstrate that the proposed framework achieves superior performance, striking a favorable balance between effectiveness and efficiency.

Outcome-based post-training methods for mathematical reasoning rely on binary feedback, treating all correct trajectories equally. However, this masks a critical distinction: some correct paths are structurally brittle and prone to cascading errors under inevitable autoregressive sampling noise, while others are robust and maintain their correctness despite these natural decoding variations. Optimizing indiscriminately over these brittle solutions leads to inefficient learning, as the model wastes capacity memorizing brittle reasoning strategies. To address this, we propose Trajectory Robustness-Driven Evolutionary Self-Training (TREST). Instead of relying on binary feedback, our framework explicitly prioritizes trajectory robustness by integrating evolutionary algorithms with supervised fine-tuning. Crucially, our analysis reveals that trajectory robustness serves as a strong natural proxy for genuine mathematical insight. By conceptualizing reasoning paths as evolving individuals within a population, this evolutionary process naturally marginalizes brittle, brute-force calculations in favor of robust, insight-driven strategies. By effectively internalizing these strategies, TREST achieves superior reasoning performance. Experiments on complex mathematical benchmarks demonstrate that under aligned budgets, TREST outperforms outcome-based baselines, establishing a highly effective alternative to conventional post-training paradigms.

Scaling robot data is crucial for building generalist Vision-Language-Action (VLA) models, yet robot trajectories are fundamentally harder to scale than web-scale image-text data because they require embodied collection and sparsely cover the physical world. This makes representation quality a central bottleneck: under a fixed robot-data budget, VLA pre-training must convert limited trajectories into transferable visual-action knowledge, rather than merely fit actions. We propose VLAct, a VLA-oriented VLM backbone trained with a representation-centric pre-training recipe. VLAct preserves the broad VLM prior, avoids over-specializing the backbone to a single action head, and encourages shared action semantics across embodiments through VLM-prior preservation, multi-head continuous action co-supervision, and a partially unified cross-embodiment action layout, while leaving downstream users free to attach task-specific action heads during fine-tuning. Across multi-embodiment simulation benchmarks, real-world robot experiments, and unseen-embodiment transfer, VLAct consistently improves downstream performance under fixed fine-tuning protocols. On LIBERO-Plus and RoboTwin 2.0, VLAct surpasses large-scale industrial VLA systems such as ABot-M0 and LingBot-VLA, achieving 82.6\% and 92.5\% success, respectively. Most notably, on RoboCasa-GR1, a humanoid embodiment never seen during pre-training, VLAct with only 20\% of downstream trajectories already outperforms the full-data GR00T-N1.6 baseline. These results are obtained using fully open-source data and only a 16-GPU training setup, showing that representation-centric pre-training is an important independent axis of VLA progress beyond data scaling. All models and training pipelines will be open-sourced.

The design of an efficient curriculum has become increasingly important for reinforcement learning with verifiable rewards (RLVR). The majority of existing curriculum strategies select training problems based on **difficulty**---typically measured by success rate---yet difficulty is an indirect proxy that conflates problem hardness with actual training utility. In this paper, we propose to drive curriculum design not by how hard a problem is, but by how much the model improves from training on it. We formalize this principle through the **improvement function** (IF), which measures the gain in a prompt's expected reward under an infinitesimal GRPO policy gradient step. Leveraging the IF framework, we formally characterize the recently discovered **edge of competence** (EoC) phenomenon: a problem is at the model's EoC when its improvement is within a constant factor of the maximum over the training set. Analyzing GRPO on $L$-step compositional reasoning tasks, we prove that an EoC-induced curriculum achieves target mastery in $\widetilde{\Theta}(\log L_{\max} / (\eta \log d))$ steps---an exponential improvement over the $\widetilde{\Theta}(L_{\max} / (\eta \log d))$ steps required by a uniform mixture of difficulties. Motivated by this theory, we propose ReCUR, a practical algorithm that maintains a stratified retry buffer of previously failed problems, resampling them until they yield positive reward or exhaust a retry budget. Experiments on multimodal reasoning benchmarks show that ReCUR consistently improves the performance of GRPO and DAPO, providing empirical evidence consistent with the improvement-driven curriculum perspective.

Downstream scores are the standard evidence for progress in point-cloud self-supervised learning (SSL). But a score is not a mechanism: it does not explain why a representation improves. We diagnose three simple questions in the evaluation pipeline: what signal is learned, what geometry is visible, and what the score measures. These questions test whether a gain comes from the pretext objective, from the geometric support available in the input, or from the downstream readout itself. Applied to cross-modal alignment, autoregressive prediction, and masked autoencoding, our diagnostics reveal that downstream scores often fail to identify the source of a gain. For Concerto, coordinate-only signals explain 28.9\% of the pretext-loss response but recover only 6.9\% of the downstream gain on ScanNet, separating pretext-loss explanation from representation transfer. For masked autoencoding, removing spatially coherent regions hurts much more than removing the same number of random points, showing that the visible geometry, not just the retained fraction, drives the score. For PointGPT, relaxing mask and order assumptions reduces transfer by at most 2.78 points on ScanObjectNN and 0.22 points on ShapeNetPart, showing that high scores can persist after weakening the intended autoregressive mechanism. These findings caution against treating downstream gains as self-evident representation progress in point-cloud SSL. We propose a diagnostic protocol that asks not only whether a score improves, but why it improves.

Generative molecular design is shaped by simple proxy benchmarks for drug-like properties and models pretrained on large pharmaceutical datasets. This combination drives strong benchmark metrics but limits transferability to domains structurally distinct from drug discovery. To overcome this limitation and drive discovery toward real, scientifically grounded targets, we introduce the Nanotechnology Molecular Optimization (NMO) Benchmark, which bridges machine learning (ML) and quantum materials science. NMO acts simultaneously as a rigorous testbed for the ML community and a discovery engine for nanotechnology research. The suite replaces proxy oracles with quantum simulations and introduces strict protocols that prioritize scientific utility over leaderboard-oriented overfitting. The physics-based NMO tasks impose hard structural constraints and rugged fitness landscapes, posing fundamentally new requirements on generative models. Notably, advanced molecular optimization methods underperform much simpler approaches on the NMO tasks. We develop a new baseline method identifying the critical components to solve the NMO tasks, including a novel representation for modeling structural constraints and a domain-agnostic pretraining strategy to eliminate pharmaceutical dataset bias. Our results surpass state-of-the-art physical properties and reveal previously unknown structural motifs, offering new insights for the nanotechnology community and demonstrating that ML can drive genuine scientific discovery.

As foundation models scale toward fusing more heterogeneous visual streams, understanding how diverse encoders interact under joint training becomes a prerequisite for principled design. Yet large vision-language models (LVLMs) currently lack the tools to do so, and parameter-efficient encoder configurations remain hard to identify before training. To re-examine encoder roles under joint training, on the 16-benchmark Cambrian-1 suite we retrain and evaluate all 31 non-empty subsets of five common vision encoders under a unified pipeline ($\approx$20k GPU-hours total), and report three findings. First, retraining each subset from scratch reveals encoder rankings that differ from those obtained by masking encoders on a fixed checkpoint, including which encoder ranks first overall. Second, we decompose each encoder's contribution into two axes, \emph{Capacity}, the score an encoder reaches on its own, and \emph{Necessity}, the drop when it is removed from the full pool. The two axes are not interchangeable. Pairing the two highest-Capacity encoders is suboptimal, while pairing a high-Capacity anchor with an adaptive complement matches the full five-encoder model. Adding further encoders beyond this pair yields only marginal gains. Third, at fixed parameter count, per-encoder pre-projector effective rank explains the residual score variation. The strongest pairs combine an anchor whose rank survives joint training with a complement whose rank \emph{expands} under it, suggesting that higher-rank, less-collapsed projector inputs correspond to a more favorable optimization regime at the encoder–projector interface. Together, the Capacity–Necessity decomposition and the pre-projector rank analysis, along with comprehensive evaluation through retraining, expose a methodological gap in multi-encoder LVLM design, and offer concrete primitives for closing it.

Unlearnable Examples (UEs) have emerged as a promising data protection strategy against unauthorized model training, which adds imperceptible perturbations into data to degrade generalization. Existing studies predominantly assume a fully trainable target model, where unlearnability is achieved by disrupting clean feature representations. However, this assumption grows increasingly unrealistic with the prevalence of pretrained models, where unauthorized users can freeze the backbone and perform linear probing, preserving discriminative representations and thereby invalidating the core mechanism of previous UEs. In this paper, we investigate this practical yet challenging linear probing setting and reveal a fundamental vulnerability. To address this, we propose $\textbf{DIVERT}$ ($\textbf{D}$ecision-boundary d$\textbf{I}$version $\textbf{V}$ia s$\textbf{E}$mantic-decoupled o$\textbf{R}$thogonal $\textbf{T}$argets), a novel framework that shifts the design principle from feature disruption to explicit decision boundary diversion. Our intuition is that projecting perturbed samples into a subspace orthogonal to the semantic manifold induces spurious linear separability, thereby steering the decision boundary away from original prototypes. Building on this principle, DIVERT first constructs synthetic class anchors within the semantic-decoupled subspace, then employs a cross-attention mechanism to jointly optimize these anchors while pushing perturbed samples toward them to establish spurious separability. Subsequently, it incorporates a surrogate linear head to simulate the linear probing procedure, further refining perturbations to actively divert the decision boundary. Extensive experiments demonstrate that DIVERT establishes effective unlearnability across diverse datasets and backbones.

Despite remarkable progress in image-text understanding, vision-language models (VLMs) still struggle with compositional reasoning. In particular, they often fail to distinguish relational direction and attribute-object binding, leading to similar representations for semantically different image-text pairs. This problem mainly stems from the reliance on global image-text alignment, which captures coarse correspondence but overlooks fine-grained compositional structures. Toward this end, we propose an evidence-aware framework, termed Directional Relation Bucketing with Binding Localization (DELTA), to improve compositional understanding in VLMs. The key idea of our DELTA is to improve compositionality of VLMs from two complementary perspectives: direction modeling and binding localization. More specifically, we first leverage a learnable gate to model the cumulative contextual distance between anchor terms, thereby assigning relation concepts to direction-aware discrete buckets. To enrich textual representations with fine-grained relational semantics, we calibrate global semantic representations by incorporating intermediate-layer hidden states. Furthermore, DELTA leverages textual cues to ground visual evidence for object attributes, pulling image patches closer to their matched attribute descriptions while pushing them away from incorrect augmented ones. Extensive experiments on four compositional datasets demonstrate the effectiveness of the proposed DELTA. Our implementation is available at https://anonymous.4open.science/r/DELTA-2D60.

Large language model agents are increasingly used as synthetic participants in social, behavioural, and decision-making research, yet it remains unclear whether they can express graded cognitive diversity rather than collapsing toward generic competence or stylised persona descriptions. We introduce a cognition-suite evaluation for LLM persona simulation, testing whether models reproduce population-level variation across structured reasoning tasks, self-report cognitive/personality measures, cross-construct associations, and downstream behavioural reasoning. Agents are conditioned on synthetic or human-derived trait profiles spanning cognitive reflection, Need for Cognition, and Big Five personality, then evaluated on CRT-style reasoning items, trait questionnaires, and Behavioural Reflection Task scenarios requiring evidence weighting, risk evaluation, and socially embedded justification. Across four matched simulation runs, agents partially recover questionnaire-style Need for Cognition and Big Five structure, especially under numeric prompting, but fail to preserve cognitive-reflection fidelity: CRT-style scores collapse toward correctness, intuitive-lure errors are almost eliminated, and reflection-related association structure weakens. Downstream BRT responses show the same pattern of flattening: open-ended decisions compress into over-regularised behavioural profiles rather than preserving human-like variation. The evaluation reframes persona simulation as a problem of cognitive-diversity fidelity: useful synthetic populations must preserve not only demographic or personality descriptors, but the structured variation in reasoning style, error type, and behaviour that those descriptors are meant to organise.

Online reinforcement learning (RL) agents increasingly depend on knowledge acquired offline to achieve practical efficiency. Originally studied in offline-to-online RL, this paradigm now spans foundation model post-training and embodied intelligence, with prior types expanding from offline datasets and pre-trained policies to increasingly diverse knowledge sources such as multimodal foundation models and generative world models. Offline priors have become central to how deep RL is developed and deployed. However, this reliance introduces a challenge that the prevailing benchmark-driven paradigm cannot resolve: because prior validity varies across deployments and shifts during training, no single approach to managing it is universally optimal, and benchmark rankings offer limited guidance for real-world deployments. Rather than pursuing universal solutions, we argue that the field should shift to diagnosis-driven tension management, in which deployment-specific evidence guides how the learner relates to its priors throughout training, enabling both flexible and adaptive deployment. We support this position with a framework characterizing how priors reshape online optimization through three functional roles, controlled experiments demonstrating help-or-hurt reversals, cross-domain evidence from foundation model post-training to embodied intelligence, and engagement with five substantive counterarguments.

Search agents based on large language models address complex information-seeking tasks by iteratively issuing queries, retrieving evidence, and integrating information. Effective training of such agents requires supervision beyond final-answer correctness, as outcome-level rewards cannot distinguish efficient evidence acquisition from redundant or misdirected search. Recent process-level methods introduce intermediate supervision, but typically apply uniform signals across trajectories. However, such uniformity overlooks the state-dependent nature of process quality, where the value of an intermediate decision depends on the current evidence state. This motivates a systematic analysis of how process quality varies across trajectories with different evidence states and final outcomes. To this end, we analyze search trajectories on BrowseComp-Plus and xBench, revealing two consistent patterns: successful trajectories often suffer from post-alignment inefficiency (redundant searches), whereas failed trajectories typically struggle with misdirected exploration and insufficient coverage. Motivated by these findings, we propose Process-Aware Grouped rEward optimization (PAGE), a framework for search agents that provides a concrete instantiation of state-dependent process supervision. PAGE partitions trajectories according to final outcomes, assigns differentiated process rewards to successful and failed groups, and integrates them with outcome supervision at either the reward or advantage level. Experiments on three backbones across six multi-hop QA and web-search benchmarks show that PAGE consistently improves answer accuracy over outcome-only baselines.

Multimodal alignment under scarce paired supervision aims to align independently pretrained unimodal encoders using limited image--text pairs, offering a practical alternative when large-scale joint pretraining is infeasible. Such scenarios call for lightweight objectives that go beyond local correspondences and capture distributional mismatch between heterogeneous embeddings. Existing methods often derive supervision from limited pairs through contrastive objectives or geometry-preserving regularization, encouraging instance-level correspondence or structural consistency but providing limited direct supervision on global cross-modal mismatch. This motivates a distribution-level objective that complements paired supervision by comparing image and text embedding distributions. In this paper, we propose $\textbf{Spec}$tral Characteristic Distribution $\textbf{Align}$ment ($\textbf{SpecAlign}$), a lightweight spectral alignment framework that reduces cross-modal mismatch by comparing image and text characteristic functions over learnable spectral queries. Since characteristic functions uniquely determine probability distributions, SpecAlign provides a principled signal for capturing multi-scale discrepancies beyond pointwise alignment. To obtain informative and stable queries, SpecAlign employs a Structured Direction--Radius Sampler (SDRS) that decouples direction and radius modeling. Extensive evaluations on transfer, retrieval, robustness, and representation analyses demonstrate consistent gains, validating spectral distribution-level supervision for data-efficient multimodal alignment.

Model merging has emerged as a promising paradigm for composing large language models directly in weight space, enabling training-free integration of specialized models. However, existing methods rely on parameter-space averaging that systematically induces generation dysregulation---a spectrum of failure modes ranging from structural repetition and termination failure to circular reasoning, and at the extreme, full semantic collapse into incoherent symbols. Even when source models exhibit near-zero dysregulation, existing methods introduce it at $>$97\% (14B/32B scales), collapsing reasoning benchmarks by up to 70 points. We propose Sparse Complementary Fusion with Reverse KL (\textbf{SCF-RKL}), a data-free merging framework that selects complementary parameters via reverse KL divergence on parameter-group proxy distributions. This structure-preserving, sparsity-inducing design maintains proxy-space geometry---theoretically motivated via entropy and subspace bounds---and empirically suppresses generation dysregulation while integrating new capabilities. Extensive experiments on 24 benchmarks across 7B--32B models spanning reasoning, instruction following, safety, and vision demonstrate that SCF-RKL achieves the best overall performance across scales while maintaining near-zero dysregulation ($<$1\%) and strong generalization.

Equipping VLM agents with world modeling capabilities has shown strong potential for complex reasoning and long-horizon planning, while reducing the dependence of policy learning on costly real-world interactions. Existing methods mainly rely on prospective simulation to predict the consequences of candidate actions. However, this forward-only paradigm focuses on *what will happen next* and provides limited constraints for verifying whether an action is causally consistent with the observed state transition, which can lead to plausible-looking but physically incoherent behaviors. In this paper, we challenge the view of world modeling as only prospective prediction and introduce Retrospective World Modeling, a new agent learning paradigm that enables agents to reason backward by estimating the retrospective attribution distribution $P(\hat{a}_{t}|s_t, s_{t+1})$ for the action that most likely caused a given transition. Based on this capability, we formulate the Self-Consistency Reward (SCR), an intrinsic signal that measures the probabilistic consistency between the policy action and the retrospective explanation. Integrating SCR into reinforcement learning provides dense transition-level feedback and steers agents toward behaviors that are both task-effective and physically grounded. Extensive experiments across diverse agentic tasks show that our method substantially improves policy robustness and generalization over prospective-only world modeling baselines.

Forensic verification has converged on a binary "real vs. fake" label that conflates fully synthetic, tampered, and AI-retouched images despite their very different consequences. We argue that GenAI manipulations decompose along two complementary channels: a camera channel of sensor fingerprints (destroyed by neural processing), and a semantic channel of scene-level coherence (disrupted by content edits). Each manipulation type leaves a distinctive two-channel signature. We instantiate this taxonomy in 2CAP (2-Channel Authenticity Protocol), pairing a contrastively-trained camera encoder with a frozen semantic encoder to serve two applications on a shared backbone: (i) four-class authenticity classification via cross-attention fusion with learnable reliability weights; and (ii) evidence generation, where an authentic image serves as a reference for a query and the two encoders produce per-channel alignment scores together with a patch-level saliency map. These signals are supplied as privileged context to a frozen Vision Language Model (VLM) through an Observe-Generate-Refine loop, yielding explanations with localization without forensic fine-tuning. On a multi-source benchmark, 2CAP attains the best classification performance, including strong retouching detection where existing methods fail, and improves explanation quality.

Stereo matchers estimate depth by comparing the left and right images from a stereo camera pair, but they assume that the two cameras are accurately calibrated and rectified. In real deployments, small physical shifts after calibration, caused by camera motion, mechanical drift, vibration, or installation changes, can break this assumption and degrade depth estimation. Online stereo rectification is therefore needed to correct the image pair during operation and recover a stereo geometry where standard stereo matching can work reliably. Recent online rectification methods mainly optimize row alignment, making the left and right views of the same scene point fall on the same image row. We show that this target, while necessary, is insufficient for deployable stereo depth estimation. A rectifier may make a small set of matched points look well row-aligned, but still distort the full images in ways that hurt stereo matching: useful image regions may be cut off, large blank areas may appear, objects may be stretched or squeezed, or image scale may change unevenly across the view. We propose VirtualRect, an online stereo rectification method designed for reliable real-world deployment. Rather than treating rectification as simply making the left and right image rows line up, this method estimates a rectified virtual stereo camera. This makes the rectified images and the camera parameters come from the same geometry, so the images used for matching and the model used for depth computation remain consistent. To improve stability, it further constrains the rectification behavior over the whole image and reduces the influence of geometrically unreliable matches. Experiments on multiple datasets show that, compared with the prior state-of-the-art online rectifier, VirtualRect reduces vertical flow error by 17\% on Carla-Flowguided and 7\% on Semi-Truck Highway.

Many decision-level fusion methods implicitly assume that semantically aligned soft outputs can be directly aggregated once they share the same label simplex. However, this assumption can fail even under identical label semantics, as oracle-specific differences in confidence sharpness, boundary uncertainty, and tail-mass allocation may place soft outputs in incompatible probability geometries and bias direct soft fusion. To address this issue, we formalize this failure mode as geometric incomparability and propose a pre-fusion comparability recovery framework, which learns a shared consensus through constrained oracle-specific rectifiers on the probability simplex. The framework includes a closed-form variant with frozen geometry descriptors and an alternating-refinement variant with consensus-dependent descriptors. Theoretically, we analyze the rectifier family, prove exact recovery under matched radial distortions, characterize the bias of direct fusion, and establish stable recovery under approximate mismatch. Experiments on controlled synthetic stress tests and real black-box decision-level settings show that comparability recovery consistently reduces fusion bias and improves reliability over direct aggregation. These results indicate that geometric comparability is a necessary condition for reliable soft-output fusion. The code is available at \url{https://anonymous.4open.science/r/CFCR-ACR-B8F1}.

Conditional diffusion models have emerged as a powerful paradigm for image restoration, leveraging pre-trained generative priors to recover high-fidelity details from degraded inputs. However, existing methods typically follow a static inference protocol, treating the degraded observation as a fixed control signal throughout sampling. This overlooks the potential of inference-time scaling, where guidance can be refined on the fly to improve output quality. Recent test-time optimization approaches that manipulate noise variables or denoising trajectories often compromise structural fidelity, while those targeting text embeddings lack the spatial precision needed for restoration. To address this, we propose Condition-Aware Test-Time Optimization (CATTO), a training-free method that iteratively refines the visual conditioning signal at inference time. CATTO performs reward-aligned condition refinement under the pre-trained prior to enhance perceptual quality. To avoid complex backpropagation, we design a gradient-free optimization strategy guided by a joint objective: maximizing a perceptual reward while enforcing trajectory- and condition-level consistency to explicitly preserve structural fidelity. This process is accelerated by optimizing within a low-dimensional frequency subspace and reusing reliable updates across nearby denoising steps. Experiments on image restoration benchmarks show that CATTO improves perceptual quality over strong diffusion-based baselines without updating model parameters.

Multi-modal image fusion (MMIF) aims to form a single image by integrating shared information, preserving complementary cues, and coordinating cross-modal conflicts across modalities. However, due to the absence of ground-truth fused images, existing MMIF supervision commonly uses spatial-domain sources or gradient variants as surrogate ground truth, making the supervision mechanism inherently misaligned with the goal of MMIF and causing pixel-level compromise or modality bias. To address this, we propose a relation-constrained supervision paradigm that moves fusion supervision from the spatial domain to a learned relation space. Instead of forcing the fused image to approximate the sources, we use frozen pretrained representation models as information providers and design a learnable feature adapter to align heterogeneous DINO and CLIP features into a unified supervision space. The adapter infers three relation parameters, namely sharedness, dominance, and coordination radius, which define three losses corresponding to the MMIF's goal. To make this space reliable, we devise a self-supervised contrastive ranking objective tailored to the adapter and couple it with the fusion network through alternating optimization. Extensive experiments show that the proposed supervision space yields significant gains regardless of which mainstream backbone the fusion network adopts, offering a supervision paradigm better aligned with the goal of MMIF. Our code will be publicly available.

Current parameter-efficient fine-tuning (PEFT) methods like Low-Rank Adaptation (LoRA) are structurally agnostic, applying uniform configurations across all layers, which overlooks their vast functional heterogeneity. We propose Structure-Aware LoRA (SA-LoRA), a new framework that automatically tailors fine-tuning intensity to each layer's intrinsic complexity. Specifically, we leverage the Stable Rank as a spectral metric to align the adaptation magnitude of each layer with its pre-trained spectral structure, enabling an automated, principled allocation of learning capacity that directly addresses the structural agnosticism of existing methods. To enhance adaptability and robustness, we introduce a hybrid calibration mechanism that fuses the task-agnostic prior with task-specific gradient feedback, underpinned by a budget-conservation principle to ensure stability. Extensive experiments demonstrate that SA-LoRA consistently outperforms strong PEFT baselines, often achieving state-of-the-art performance with enhanced stability. The code is available at https://anonymous.4open.science/r/SA-LORA-6809.

Emerging super-nodes tightly couple multiple servers with symmetric high-bandwidth fabrics, substantially weakening the legacy bandwidth cliff between nodes. This makes cross-boundary TP/EP a viable part of the strategy space, shifting the optimal strategy basin away from the legacy practice of keeping TP/EP within a node. Yet adapting the parallel strategy to new platforms still requires expensive profiling and brittle manual tuning. We propose Symbolic Barrier-aware Planner (SBP), a profiling-free planner that selects DP/MP/PP factorization and activation checkpoint strength by computing a barrier-aware symbolic score from model shapes, collective semantics, and hardware specifications. On legacy clusters, SBP recovers measured-best configurations; on super-nodes, it captures the regime shift and achieves $1.40\times$ step-time speedup and a 10.62 percentage-point MFU improvement over the best node-local Megatron-style baseline on 128-die training. SBP explores the strategy space in seconds on a single CPU, without accelerator profiling for strategy selection.

Video Reasoning Segmentation (VRS) aims to infer and segment target objects specified by reasoning queries in videos. Existing methods typically first infer the target and generate a few segmentation tokens for it, which are then fed into a mask decoder for mask prediction. However, such segmentation tokens tends to be both semantically monotonous and spatially ambiguous, making them largely insufficient to represent the target to guide mask prediction. In this paper, we propose an explicit visual object grounding framework for VRS. Specifically, we first propose a spatial-aware prompt generation and refinement scheme, in which we query a multimodal large language model to generate explicit spatial prompts (\ie, bounding box and point set) for the target and then refine them through multi-step dialogue. Furthermore, we introduce a query diversification and alignment module to generate auxiliary queries that describe the same target from different perspectives. We then enforce cross-query consistency between their predicted spatial prompts to alleviate the training bias caused by query limitation. Finally, we selects several reliable frames using a prompt-based keyframe selection strategy to support mask decoding and propagation. Extensive experiments on eight datasets demonstrate that our method significantly outperforms state of the art methods.

Existing hallucination detectors for retrieval-augmented generation (RAG), whether based on model outputs (e.g., likelihood) or internal representations, typically apply a uniform detection strategy across responses. However, we find that different detection signals are informative in different response regimes, making a uniform strategy insufficient for both short- and long-form generation. In short-form tasks such as factoid QA, hallucination typically appears as selecting an incorrect final answer among multiple context-relevant candidates. In this regime, model-output-based signals that discriminate among plausible answers are particularly important. We therefore propose a bidirectional likelihood measure that evaluates logical consistency based on the retrieved evidence and a reasoning step generated after the response. In contrast, long-form descriptive responses are more prone to gradual drift: as generation proceeds, the model increasingly relies on its own prior generations and parametric knowledge, while unsupported continuations may remain locally plausible, making output-based cues less informative. For this regime, we measure context--knowledge conflict by tracking the directional alignment of context-grounding versus internal-knowledge contributions in the hidden states. Building on these regime-specific signals, we introduce ARGUS, an adaptive hallucination detector that emphasizes the more effective signal according to the response regime. Experiments on multiple RAG benchmarks show that our method achieves strong performance across both short- and long-form settings.

Understanding long-form video remains a fundamental challenge for multimodal large language models (MLLMs). Sparse frame sampling fails to capture fine-grained visual details, while dense sampling quickly exceeds context length limits. Retrieval-augmented generation (RAG) offers a promising middle ground by selectively retrieving relevant video segments for grounded generation, yet its effectiveness critically depends on the quality of the video segments used as retrieval units. In this paper, we investigate RAG for movie understanding, which demands story-level reasoning over characters, events, and narrative arcs spanning hours of content. Scene segmentation, a long-studied problem that partitions movies into semantically coherent units, is a natural candidate for defining such retrieval units. We reexamine whether existing methods actually serve this role through comprehensive evaluation on downstream movie understanding tasks, and find that they consistently fail to outperform naive uniform temporal chunking. Our audit of the most standard scene segmentation benchmarks reveals why: current annotations prioritize visually salient transitions over narrative event structure. Motivated by this mismatch, we introduce NarraScene, a narrative-centric scene segmentation dataset annotated with a three-level cognitive taxonomy spanning physical, character, and narrative change, where every valid boundary requires a narrative-level shift. When used as retrieval units, these narrative-grounded segments outperform uniform chunking on downstream movie understanding tasks, suggesting that the central challenge for scene segmentation in movie RAG is not detecting boundaries, but identifying the narrative event units that matter for movie understanding.

Large Language Models (LLMs) are increasingly deployed as automated judges to scale supervision for data curation, reinforcement learning (RL), and agentic systems. While existing works have extensively explored bias in pretrained LLMs, the origins of such inherent biases remain largely untraced. We trace these biased tendencies to cognitively biased patterns (e.g., Authority, Bandwagon) latent in training corpora, and expose that such patterns are naturally prevalent in large-scale pretraining corpora yet remain entirely undetectable by existing data cleaning pipelines. To demonstrate that this underexplored threat can be deliberately exploited, we introduce BiasTrojan, a framework that concentrates and injects these naturally occurring bias patterns into training samples via context-aware bias cues and contrastive preference pairs, augmented with counterfeited reasoning chains for efficient injection. Experiments across several LLMs from 7B to 70B on human-preference and fact-related datasets show that mere hundreds of deliberately biased samples suffice to compromise LLMs into biased evaluators, overriding their factual knowledge. The injected biases generalize robustly out-of-domain and persist despite massive continual post-training. These findings reveal that this underexplored latent threat poses far greater risks than commonly recognized: biased LLM judges whose evaluations propagate irreversibly downstream, underscoring the critical need for bias-aware auditing and strict scrutiny of LLM training data.

Discovering hit molecules requires not just high binding affinity, but also the identification of diverse binding modes, which are critical for experimental assays. Existing generative approaches have predominantly relied on optimizing a scalar docking score, obscuring the distinct contributions of key binding determinants. To this end, we introduce a paradigm shift by formulating target-based drug design as a multi-objective exploration task, where each objective explicitly corresponds to enhancing interactions with a specific hotspot. Here, we introduce BindMol, a novel generative framework driven by a customized multi-objective reinforcement learning algorithm. By incorporating explorative preferences during training, our approach efficiently uncovers molecules with diverse and desirable binding profiles. Empirical results demonstrate that BindMol facilitates the discovery of high-affinity compounds characterized by both structural novelty and diverse binding modes. Validated across target-based drug discovery and multi-property optimization tasks, our approach provides a versatile paradigm for goal-oriented drug discovery.

Vision-language models (VLMs), such as CLIP, have shown remarkable transferability in the biomedical domain, and prompt tuning enables efficient adaptation with limited supervision. In practice, biomedical images exhibit pronounced acquisition-induced style shifts, especially across imaging sites and protocols, while cross-modality transfer further introduces compound shifts from different physical acquisition processes. However, existing prompt-tuning methods ignore acquisition-induced style shifts, making the prompts sensitive to source-specific style and limiting the transferability to unseen medical domains. To address this, we formulate acquisition-induced style shifts as mixture shifts, where each domain is viewed as a mixture of latent style components. Under this formulation, the target risk depends on both mixture weights and latent component risks, motivating prompt adaptation that accounts for style-specific variation rather than relying on a single source-specific prompt. In this paper, we propose SaMoE, a Style-aware Mixture-of-Experts Multimodal Prompt Learning framework for adapting VLMs to biomedical domains under acquisition-induced style shifts. To account for latent component risks in prompt adaptation, SaMoE represents multimodal prompts with latent style-conditioned prompt transformations and composes them through image-conditioned routing, enabling each image to induce style-adapted visual prompts. By extracting style cues from intermediate visual representations, the sparse expert composition adapts style-related prompt weights without domain labels and is injected across multiple VLM layers to align with hierarchical visual-language representations. Extensive experiments on 15 medical datasets across 12 modalities and 10 organs demonstrate significant improvements in both accuracy and generalizability over state-of-the-art methods.

3D Gaussian Splatting (3DGS) has become a standard representation for real-time novel-view synthesis. Yet, its quality--efficiency trade-off still relies heavily on adaptive density control. Existing cloning, splitting, and pruning rules use proxy signals—such as position gradients, opacity, image-space error. Although effective in practice, these signals do not directly answer the structural question behind density control: where should new Gaussians be allocated to reduce the multi-view reconstruction loss, and which existing primitives can be pruned with minimal penalty? To address this, we propose first-variation birth–death control for 3DGS, a principled approach that replaces heuristic structural decisions with a variational score derived from the reconstruction loss. At each structural update, we freeze the current Gaussian population, attach an auxiliary mass coordinates to each primitive, and differentiate the multiview loss with respect to these coordinates. We demonstrate that this score is a coordinate of the first variation of Splat Regression Model, giving a rigorous local descent surrogate for birth-death moves: high-score Gaussians are selected as birth sites, low-score low-opacity Gaussians are death. Across standard 3DGS benchmarks, integrating this score into 3DGS-MCMC yields state-of-the-art reconstruction quality and outperforms existing methods, delivering the most significant gains under tight Gaussian budgets. Ultimately, our work provides compelling evidence that gradient-based variational metrics should replace proxy heuristics for structural updates in 3DGS.

Diffusion models have shown promise in learning to solve constraint optimization problems. However, they are mostly restricted to problems with binary variables and rely on graph neural networks, hindering their application to a broader range of problems such as those with general discrete variables or constraint structures that necessitate global rather than local reasoning. We investigate the use of Diffusion Transformers to address the aforementioned limitations. A naive implementation performs poorly due to a fundamental mismatch between the standard diffusion process and constraint solving: while the former applies small, incremental denoising across all variables, the latter requires substantially altering specific subsets of variables to attain feasibility or optimality. Our method, Blocked Gibbs Diffusion Transformer (BloGDiT), is the first to address this limitation by replacing standard joint Gaussian denoising with Blocked Gaussian denoising. BloGDiT uses iterative block resampling and anneals the block size over time to facilitate large, targeted edits within a block of variables. Across Sudoku, Graph Coloring, Maximum Independent Set, and MaxCut, BloGDiT matches or outperforms existing methods, demonstrating that blocked Gibbs-style diffusion provides a highly effective inductive bias for Transformer-based constraint satisfaction and optimization.

Modern large language models increasingly require long contexts for reasoning and multi-document tasks, but attention's quadratic complexity creates a severe computational bottleneck. We present *Block Sparse Flash Attention* (BSFA), a drop-in replacement that accelerates long-context inference while preserving model quality. Unlike methods that predict importance before computing scores, BSFA computes exact query-key similarities to select the top-$k$ most important value blocks for each query. By comparing per-block maximum scores against calibrated thresholds, we skip approximately 50% of the computation and memory transfers for pruned blocks. Our training-free approach requires only a one-time threshold calibration on a small dataset to learn the per-layer and per-head attention score distributions. We provide a CUDA kernel implementation that can be used as a drop-in replacement for FlashAttention. On Llama-3.1-8B, BSFA achieves up to $1.13\times$ end-to-end speedup on LongBench with only a $1.1\%$ accuracy drop, and up to $1.24\times$ on Needle-in-a-Haystack retrieval at a $1\%$ accuracy drop. The attention kernel itself accelerates by up to $1.38\times$. We compare BSFA against five recent sparse attention baselines (SpargeAttention, MInference, FlexPrefill, XAttention, and BLASST), and verify the method on Qwen2.5-7B and on A6000 and H100 GPUs. The implementation is available at https://github.com/Anonymous44414/Block-Sparse-Flash-Attention.

Novel View Synthesis(NVS) for thermal infrared scenarios has become increasingly important in practical applications. Recent advances in 3D Gaussian Splatting(3DGS) have demonstrated its strong performance for NVS, and a handful of preliminary efforts have explored its extension to thermal NVS. However, existing 3DGS-based methods for thermal NVS often suffer from severe local blur in rendered images. Theoretical and empirical analyses reveal that the single-channel nature of thermal data allows intensity discrepancy to be compensated through opacity adjustment, which can erroneously amplify the opacity of floating Gaussians, especially when the background behind them is insufficiently represented. To address this issue, we propose a method termed Floating Gaussian Suppression (FGS) for thermal NVS. Specifically, we leverage an over-saturation of luminance to introduce an implicit gradient penalty, which guides the optimizer to suppress the opacity of floaters to match the ground truth. By effectively reducing the opacity of these floaters, this mechanism alleviates the occlusion of distant details, thereby mitigating local blur. Moreover, to preserve rendering fidelity, the luminance parameters of Gaussians are optimized concurrently. Experimental results demonstrate that our method effectively mitigates local blur and produces sharper and more faithful thermal reconstruction compared with the baseline methods.

Large language models (LLMs) excel at reasoning across diverse domains, from mathematical problem-solving to code generation, yet struggle with strategic reasoning that requires long-term planning, opponent modeling and integrating tactical calculation with strategic understanding. Board games, with their fully observable states, perfect-information environments and objectively verifiable decisions, offer ideal testbeds for probing these capabilities. We introduce Board Game Arena (BGA), a multi-dimensional benchmark that evaluates LLMs' strategic reasoning along three cognitive axes—deductive, inductive, and abductive reasoning—through three challenging task types across five classic board games, yielding 45 sub-datasets with over one million samples. We evaluate 28 state-of-the-art LLMs on BGA and find that current models exhibit fundamental limitations in strategic reasoning for board games. While models demonstrate competence in understanding game rules and identifying legal moves, they struggle significantly with selecting optimal moves and applying strategic principles or tactical analysis. These results position BGA as a challenging and diagnostic benchmark for tracking progress in LLMs' strategic cognition.

Graph contrastive learning constructs augmented views and aligns corresponding instances across views to learn view-invariant representations. However, existing methods usually enforce cross-view consistency within a single representation space, lacking an explicit characterization of view-specific information. This may suppress task-relevant complementary information, reduce embedding diversity, and induce dimensional collapse. To this end, we propose Manifold-Guided Representation Disentanglement (MGRD), a plug-and-play framework for boosting graph contrastive learning. MGRD decomposes graph representations into shared and complementary subspaces. The shared subspace captures view-invariant semantics while preserving local graph structure through a graph-anchored manifold prior, whereas the complementary subspace captures residual view-specific information beyond the shared representation and is regularized by asymmetric consistency and subspace decorrelation. By jointly exploiting shared and complementary representations, MGRD balances cross-view invariance and representation diversity. Experiments on node-level and graph-level benchmarks show that MGRD consistently improves multiple GCL backbones, while theoretical and empirical analyses demonstrate its effectiveness in mitigating dimensional collapse.

Phase transitions occur when a physical system undergoes a dramatic transformation as it crosses a transition. We study whether regression methods trained only in one phase can predict the physical properties that characterize the unobserved phase. Our test sets consist of the analytically tractable Heisenberg model and angle-resolved photoemission spectroscopy measurements of a venerable high transition temperature superconductor. We find that polynomial regression and kernel methods recover the unobserved phase structure, while standard neural networks fail. To bound the error, we adopt the generalized eigenvalue problem (GEVP) into an upper bound on the worst-case transfer coefficient, defined as the ratio of mean squared error (MSE) loss on the extrapolation region to the MSE loss on the training region. Empirically, we find that our GEVP bound stays approximately constant with respect to the empirical transfer coefficient across training-set sizes and distance between the training and extrapolation region for both datasets, providing a principled framework for studying extrapolation feasibility.

Object-level spatial-temporal understanding is essential for video question answering, yet existing multimodal large language models (MLLMs) encode frames holistically and lack explicit mechanisms for fine-grained object grounding. Recent work addresses this by serializing bounding box coordinates as text tokens, but this text-coordinate paradigm suffers from a fundamental modality mismatch: object information is inherently visual, yet encoding it as text incurs a high token cost that forces aggressive temporal downsampling. We propose BoxTuning, an object-aware visual prompting framework for multimodal model fine-tuning that moves spatial-temporal object states from text-coordinate serialization into the visual stream while retaining only object identity in a compact text legend. Colored bounding boxes and trajectory trails encode geometry and motion on video frames, while the color-to-object legend provides the minimal textual identity bridge. This reduces the token cost significantly, achieving 87-93\% text token reduction in practice. It also preserves full temporal resolution, where the trajectory trails further encode inter-frame motion direction and speed within each keyframe, recovering fine-grained dynamics that text-coordinate methods are forced to discard. Experimental results on five video QA benchmarks (CLEVRER, Perception Test, STAR, NExT-QA, IntentQA) show that BoxTuning surpasses text-coordinate baselines on spatially oriented tasks and avoids their degradation on reasoning-centric tasks, establishing object-aware visual prompting as a natural and efficient way to convey object information to MLLMs.

Graph neural networks for brain connectomics treat every connection as topologically equivalent, ignoring the brain-economy trade-off between the metabolic cost of maintaining a connection and the topological value it delivers to the network. We argue that this trade-off is a property of the channel (i.e., edge) between two brain regions, not of the regions themselves, and should therefore shape how regions interact during attention computation rather than being appended as a node feature or auxiliary loss. We introduce the ecospace, a two-dimensional coordinate system that characterizes each brain connection by its economy—the trade-off between connectivity strength and anatomical cost—and uses it to guide how brain regions attend to one another. We further introduce EcoSpace Rotary Encoding (ESRE), an attention mechanism that injects the ecospace dimensions through asymmetric rotations of the query-key subspace, guaranteeing by mathematical identity that the attention score between two brain regions depends on the economy of the edge connecting them. Together, these two contributions form the Brain Graph Transformer with EcoSpace Rotary Encoding (BGT-ESRE). On brain graph classification, BGT-ESRE outperforms classical, general-purpose, and brain-graph-specific baselines by a substantial margin in both accuracy and AUC. Ablations confirm that the rotary mechanism and the brain economy-aligned transformation each contribute independently, and that asymmetric rotary injection is superior to additive bias injection of the same economy measure.

Sparse tuning is widely used to adapt large language models due to its parameter efficiency. However, its effectiveness depends not only on the number of trainable parameters, but also on the feature-interaction topology induced by the sparse update. Limited or uneven cross-dimensional interactions may isolate subsets of feature dimensions, forming information islands that hinder task-specific information mixing. To address this issue, we propose Small-World Induced Fine-Tuning (SWIFT), a sparse tuning framework that induces small-world connectivity in parameter space. SWIFT combines block-diagonal sparse updates for local feature interactions with a fixed sufficiently scattering permutation that routes these updates to distant feature groups, creating sparse long-range shortcuts without additional trainable parameters. We further show theoretically that random or block-diagonal sparse tuning can suffer from information islands, whereas SWIFT restores positive expansion through permutation-induced routing. Extensive experiments on commonsense reasoning, natural language generation, and image classification demonstrate that SWIFT consistently outperforms competitive PEFT baselines and can match or surpass full fine-tuning.

Self-supervised learning (SSL) is increasingly applied to scientific imaging domains, such as microscopy, where labels are scarce but noisy data is abundant. These domains exhibit significant signal-dependent noise, which can create a representational shortcut: since both augmented views derive from the same noisy image, the encoder can boost view agreement by also encoding noise patterns rather than semantic content. To enable noise-robust representation learning without knowledge of underlying noise models, we introduce Self-Aligned Noise Augmentation (SEANA), a drop-in module that generates noise-aware SSL views. SEANA learns an invertible variance-stabilizing transform (VST) from noisy data, then estimates the clean signal and resamples independent noise in learned VST space, requiring no clean targets, no dataset-specific denoiser training or use at inference, and no changes to the SSL objective or encoder. On CIFAR-10 and ImageNet-100 under synthetic Gaussian, Poisson–Gaussian, and multiplicative noise, SEANA improves clean-test linear-probe accuracy across contrastive/non-contrastive SSL methods, indicating higher-quality representations. On real fluorescence microscopy, SEANA improves Jurkat cell-cycle stage classification by up to 28.6pp over standard SSL and 27.5pp over denoiser-preprocessed baselines. In both settings, SEANA outperforms the denoiser-preprocessed SSL pipeline, demonstrating that learned VST-space noise-aligned resampling yields more robust representations than denoising alone. Code will be made publicly available.

High-performance text-to-image~(T2I) diffusion models suffer from diversity degradation. In this study, we attribute this challenge to a fundamental temporal mismatch: while visual features evolve dynamically in a coarse-to-fine manner in the denoising process, the text condition remains entirely static. Through empirical analysis, we demonstrate that decoupling the text prompt to prioritize low-frequency components in the early denoising stage better aligns with the visual denoising process, while simultaneously enhancing generation diversity. Driven by this insight, we introduce Dynamic Semantic Interpolation (DSI), a training-free strategy that utilizes low-frequency semantics during the initial stages to foster the sampling space and progressively recovers the full semantics, ensuring text-to-image alignment. We also provide a rigorous theoretical justification for DSI from the perspective of conditional entropy, explaining its ability to maintain a broader early sampling space. Extensive experiments across diverse state-of-the-art models demonstrate that DSI significantly boosts generation diversity while preserving text alignment and aesthetic quality.

Flicker-banding (FB), arises from temporal aliasing between a camera's rolling shutter and a display's brightness modulation, degrading screen-captured image readability with color shifts and jagged patterns. Existing single-frame methods with simplified parametric stripe models cannot reliably distinguish these artifacts from genuine texture. To address this, we conduct a systematic analysis of complex FB morphologies and reveal their significant variation across exposure settings, motivating a multi-frame bracketed RAW restoration paradigm. We construct Bricker, a synthetic–real bracketed RAW dataset built via ray-tracing-based physical simulation and automated multi-exposure capture tool. We further propose BRACE: Bracketed RAW Flicker-Banding Removal, a multi-frame restoration model that utilizes frequency-aware banding prior and a multi-scale spatial cross-attention modulator (MSCAM) for cross-exposure spatial fusion. We also introduce the Stripe Frequency Consistency (SFC) metric to evaluate banding removal. Extensive experiments demonstrate state-of-the-art performance on both synthetic and real benchmarks. Our dataset and code will be publicly released.

Offline reinforcement learning (RL) offers a promising framework for deploying autonomous systems in safety-critical settings without the risks of online exploration. However, learning policies that simultaneously achieve high performance and strong safety guarantees from fixed datasets remains a fundamental challenge. Many existing safe offline RL approaches typically rely on soft constraint formulations, which may permit safety violations and are sensitive to distributional shift. In contrast, formal methods such as Hamilton–Jacobi (HJ) reachability and Control Barrier Functions (CBFs) provide rigorous safety guarantees, but often yield overly conservative solutions, often neglecting performance. In this work, we bridge this gap by formulating safe offline RL as a state-constrained optimal control problem, where safety is enforced through hard state constraints and performance is captured via a reward function. The resulting value function satisfies a Hamilton–Jacobi–Bellman (HJB) equation, which we approximate using offline RL on fixed datasets. This formulation enables principled integration of safety guarantees with data-driven policy optimization. Empirically, across safety-critical benchmarks including boat navigation and Safety-Gymnasium tasks, our approach achieves competitive returns while exhibiting near-zero constraint violations, demonstrating a favorable balance between safety and performance in the offline setting.

End-to-end autonomous driving aims to generate safe and plausible planning policies from raw sensor inputs, and constructing an effective scene representation is a critical challenge. Driving world models have shown great potential in learning rich representations by predicting the future evolution of a driving scene. However, existing driving world models primarily focus on visual scene representation, and motion representation is not explicitly designed to be planner-shared and inheritable, leaving a schism between the optimization of visual scene generation and the requirements of precise motion planning. We present WorldDrive, a holistic framework that couples scene generation and real-time planning via unifying vision and motion representation. We first introduce a Trajectory-aware Driving World Model, which conditions on a trajectory vocabulary to enforce consistency between visual dynamics and motion intentions, enabling the generation of diverse and plausible future scenes conditioned on a specific trajectory. We transfer the vision and motion encoders to a downstream Multi-modal Planner, ensuring the driving policy operates on mature representations pre-optimized by scene generation. A simple interaction between motion representation, visual representation, and ego status can generate high-quality, multi-modal trajectories. Furthermore, to exploit the world model’s foresight, we propose a Future-aware Rewarder, which distills future latent representation from the frozen world model to evaluate and select optimal trajectories in real-time. Extensive experiments demonstrate that WorldDrive achieves leading planning performance among vision-only methods while maintaining high-fidelity action-controlled video generation capabilities. Our code and model will be made publicly available.

Multiple instance learning (MIL) is the dominant framework for whole-slide image analysis in computational pathology, typically combining a frozen patch encoder, a projection layer, and a slide-level aggregator. While encoders and aggregators have been extensively studied, the projection layer remains a largely morphology-only bottleneck. This limits endpoints such as biomarker status and survival, which are governed by a molecular state that is not fully captured by H&amp;E morphology. We introduce Molecularly Informed Staining Transform (MIST), a plug-in replacement for the MIL projection layer that uses paired spatial transcriptomics only during training to construct virtual molecular stains. MIST clusters gene expression profiles into cross-modal prototypes, anchors them in the frozen foundation model feature space, and uses them to reorganize H&amp;E patch features along molecularly guided axes. It requires no transcriptomics at inference and can be inserted before standard MIL aggregators. We evaluate MIST across 23 downstream tasks and 8 MIL aggregators. MIST improves 240 of 256 configurations over the standard projection layer, with an average gain of +3.5\%, observed consistently across endpoint types: +5.2\% on survival prediction, +3.3\% on tissue subtyping, and +2.6\% on biomarker prediction. Ablations confirm that gene-derived prototypes are the primary source of the gains, while spatial, biological, and pathological analyses show that cross-modal prototype affinities capture spatially coherent molecular programs from H&amp;E alone.

Existing evaluation methods largely rely on clean, static benchmarks, which can overestimate true model performance by failing to capture the noise and variability inherent in real-world user inputs. This is especially true for language models, which can face human-generated text queries containing mistakes, typos, or alternative ways of phrasing the same question. In this work, we introduce a theoretical framework for quantifying model sensitivity to prompt variants, or brittleness, that can enable us to disentangle data-induced difficulty from prompt-related variability. Using this framework, we design a novel evaluation pipeline, Brittlebench, to holistically evaluate the sensitivity of frontier models. We apply semantics-preserving perturbations to a suite of popular benchmarks, and observe model performance to degrade as much as ~21%. However, these perturbations do not affect all models equally: even a single perturbation alters the relative ranking of models in 63% of cases, impacting conclusions about comparative model performance. Decomposing the total variance of both state-of-the-art open-weight and commercial models, we find that semantics-preserving input perturbations can account for up to half of the performance variance for a given model. Brittlebench highlights the need for more robust evaluations and models, and allows us to systematically understand model brittleness.

Token-level distillation assumes a shared categorical support, an assumption violated when teacher and student tokenizers segment the same text differently. Existing cross-tokenizer methods sidestep this mismatch by aligning token spaces, matching likelihoods along tokenizer-specific paths, or filtering shared spans, all of which can approximate the original distributional objective or discard probability mass. We argue that decoded UTF-8 bytes are common observable events shared by every text tokenizer, while token boundaries are tokenizer-specific latent transitions. Building on this view, we introduce \emph{ByteDistill}, which compares the conditional distribution of the next observable byte rather than aligning token identities. Its loss, \emph{Byte-Level Distribution Alignment} (BLDA), projects each native softmax into a 256-way observable next-byte distribution inside aligned decoded chunks, marginalizing token-ending mass as a latent boundary transition along the observed tokenization path. ByteDistill consistently outperforms supervised fine-tuning and prior cross-tokenizer baselines in both off-policy and on-policy settings, with off-policy training improving GSM8K accuracy by $1.75$ percentage points over the current state of the art.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities, yet they frequently struggle with fine-grained visual perception, often suffering from hallucinations when faced with visually similar but semantically distinct instances. A major underlying cause is the conventional single-sample Supervised Fine-Tuning (SFT) paradigm, which isolates training instances and limits the model's ability to explicitly learn subtle discriminative boundaries. To address this, we propose C2FT, a novel Confuse-then-Contrast Fine-Tuning framework that enhances fine-grained perception by shifting from independent supervision to joint supervision. Specifically, C2FT assesses the MLLM's internal uncertainty to dynamically mine model-specific semantic confusions, subsequently constructing challenging multi-image groups comprising hard positive and hard negative samples. By interleaving these samples into a unified prompt, our approach forces the MLLM's internal attention mechanisms to cross-reference images and explicitly capture localized visual discrepancies. Extensive experiments on both Fine-Grained Visual Classification (FGVC) and visual question answering (VQA) benchmarks demonstrate the effectiveness of C2FT. For instance, on the Qwen3-VL-4B model, our approach achieves an average improvements of 3.87\% and 4.35\% points over standard SFT and GRPO baselines, respectively.

Perceiving and remembering the visual world is fundamental to navigating and interacting with our environment. Current 4D foundation models, such as camera-controllable video models or 4D reconstruction models, can perceive and reconstruct dynamic environments, but how well they remember what they have perceived remains an open question. Existing benchmarks largely rely on pixel-level metrics and lack ground truth for objects once they leave the field of view, making them unable to evaluate visual memory in an object-centric manner against references. To fill this gap, we introduce PersistBench, a dataset and metric suite that leverages 360° videos as omniscient ground truth and proposes three evaluation aspects: object permanence, motion continuity, and appearance preservation. Evaluating various models across diverse categories reveals that current models can only maintain short-term consistency that degrades significantly once objects leave the field of view. Our findings highlight the gap between current model capabilities and robust visual memory, providing guidance for future development of 4D foundation models.

Scientific AI agents increasingly retrieve evidence, reason across sources, and synthesize conclusions used in consequential decisions. Yet, their ability to do so in high-stakes domains such as health remains unclear. We introduce SciConBench, a large-scale live benchmark of 9.11K questions and expert-written conclusions from systematic reviews to evaluate open-domain scientific conclusion synthesis. The benchmark draws on an expert-validated automated evaluation pipeline that decomposes conclusions into atomic facts and measures correctness and comprehensiveness via factual precision and recall. To mitigate data leakage, we further introduce SciConHarness, a clean-room evaluation harness that equips agents with controlled web interaction to ensure valid measurement. Evaluating 8 frontier models and deep research agents, we find that factual quality remains low: under clean-room settings, the best agent achieves only a factual F1 of 0.337. Our clean-room setting consistently reduces performance relative to unconstrained evaluation, suggesting that leakage inflates estimates of models' true synthesis capabilities. Finally, we audit consumer-facing agents (e.g., Google AI Overview, OpenEvidence) and find they frequently generate incomplete and sometimes contradictory conclusions, even when the ground-truth answer is available. Overall, our results show that reliable synthesis of scientific conclusions remains an open challenge, and that clean-room evaluation is essential for assessing open-domain AI agents.

Language models (LMs) raise an alluring alternative to vector-based retrieval: \emph{generating} a relevant answer, entirely based on an in-context corpus. In spite of this appeal, prior studies consider proprietary systems or the smaller-scale reranking task, leaving corpus-scale in-context retrieval largely unexplored. In this work, we present the first systematic study of in-context retrieval on two scales practical retrievers demand: \emph{million-token} corpora and \emph{length-generalization} far beyond training-time sizes. We first introduce BlockSearch, a 0.6B LCLM retriever whose architectural and training modifications improve over prior LM baselines and length-generalize up to $10\times$ beyond its training regime. However, its retrieval still collapses under more extreme extrapolation. We trace this failure to an \emph{attention dilution} effect: as the corpus grows, irrelevant documents dominate the softmax denominator, minimizing the normalized mass on the gold document even as its pre-softmax score stays high. Motivated by this analysis, we introduce \emph{length-aware} adjustments to the attention softmax and \emph{document-level sparse attention}, improving retrieval at million-token scale, better performance than the concurrent $7\times$-larger MSA system, and comparable performance to dense retrieval. Together, our results position in-context retrieval as a viable alternative to classical retrieval, emphasizing attention control under extreme context growth as a new challenge.

Large Language Models (LLMs) have made remarkable progress on solving competition-level mathematics, yet their ability to verifying natural-language mathematical proofs remains relatively underexplored. Current proof verification primarily relies on expert inspection that is costly to scale and Olympiad-level problems represent a significant challenge in this area, as they require the meticulous evaluation of every claim in the reasoning chain. In this paper, we present a systematic study of LLM-based verification in this setting. Specifically, we introduce a human-curated dataset of 48 USAMO 2025 candidate solutions with expert grading, conduct a broad study of existing LLM verifiers in Olympiad mathematics across frontier models. We also propose an iterative self-critique pipeline TROJAN to generates high-fidelity adversarial proofs at scale. Our evaluation spans 20 models on 7 metrics across two prompt styles with three inference methods and three prompt templates, illustrating that the state-of-the-art models still have more room for improvement. To address this gap, we propose MABGrader, a Multi-Armed Bandit framework that reframes grading as arm selection over the discrete score space. MABGrader outperforms the SOTA ProofGrader by 26.4% in Quadratic Weighted Kappa and reduces grading error by 14%.

Passive image provenance asks whether pixels alone can reveal where an image came from: a human, an aggregate AI class, or a particular generator. This becomes a robustness problem once a source image can be edited before the verifier sees it. We study the problem as source--target verification under adversarial distribution shift. Our first result gives the exact best-case limit for any image-only verifier: the largest robust target-acceptance gap equals the minimum total-variation distance between the target distribution and the set of attacked source distributions. This quantity depends on the source, target, and edit class, not on the verifier architecture. Our second result explains why deployed public verifiers can fail before this statistical limit is reached. If the verifier can be emulated on the attack region to error $\varepsilon$, then a surrogate black-box attack reaches target acceptance within $2\varepsilon$ plus optimization error of the white-box optimum; score-revealing logistic and softmax heads over public features are identifiable, and approximate score access gives stable recovery bounds. Experiments on same-prompt real/diffusion benchmarks support this separation. Public CLIP-based interfaces collapse under targeted attacks, and stronger clean or adversarial retraining does not restore positive operational separation. In our evaluated settings, positive empirical upper bounds on the robust gap appear only under private low-bandwidth binary interfaces with abstention. The main lesson is simple: robust passive provenance requires both a source--target statistical analysis and an interface-aware evaluation of what the verifier reveals.

Protein-molecule virtual screening is increasingly cast as a problem of representation learning in a shared embedding space. Existing methods rely on dense holistic alignment, entangling invariant binding determinants with nuisance correlations and limiting transfer to new targets. It has been noted that binding in protein–molecule systems involves sparse cross-modality interactions: binding is governed by a small contact interface and a few decisive local interactions (e.g., hydrogen bonds, hydrophobic contacts, and salt bridges) rather than the global structures of the protein and molecule. We hypothesize that uncovering and leveraging sparse interaction patterns is critical for generalization beyond the training data, as these patterns are reusable and expected to improve performance across different scenarios. In this paper, we aim to identify and leverage sparse interaction patterns, and verify our hypothesis. Since the training data contain only observed binding pairs, we formalize this prior via a V-structure causal model under Heckman-style selection, and establish three theoretical results: (i) the latent concepts of interacting proteins and molecules are not identifiable without appropriate sparsity constraints;(ii) these concepts and their sparse interactions are component-wise identifiable under structural sparsity conditions; and (iii) a low-rank relaxation of these conditions yields subspace identifiability of the concepts and interactions. Inspired by these principles, we propose CausalBind with three implementation variants. Extensive experiments on DUD-E and LIT-PCBA benchmarks show that all variants consistently outperform strong baselines across all metrics, respectively.

Causal discovery from time series is a fundamental task in machine learning. However, its widespread adoption is hindered by a reliance on untestable causal assumptions and by the lack of robustness-oriented evaluation in existing benchmarks. To address these challenges, we propose CausalCompass, a flexible and extensible benchmark framework designed to assess the robustness of time-series causal discovery (TSCD) methods under violations of modeling assumptions. To demonstrate the practical utility of CausalCompass, we conduct extensive benchmarking of representative TSCD algorithms across eight assumption-violation scenarios. Our experimental results indicate that no single method consistently attains optimal performance across all settings. Nevertheless, the methods exhibiting superior overall performance across diverse scenarios are almost invariably deep learning-based approaches. We further provide hyperparameter sensitivity analyses to deepen the understanding of these findings. We additionally conduct ablation experiments to explain the strong performance of deep learning-based methods under assumption violations. We also find, somewhat surprisingly, that NTS-NOTEARS relies heavily on standardized preprocessing in practice, performing poorly in the vanilla setting but exhibiting strong performance after standardization. Finally, our work aims to provide a comprehensive and systematic evaluation of TSCD methods under assumption violations, thereby facilitating their broader adoption in real-world applications. The user-friendly implementation, documentation and datasets are available at https://anonymous.4open.science/r/CausalCompass-anonymous-5B4F/.

There are many causal discovery algorithms, but few guiding principles for choosing between them. Methods are usually compared through three lenses: identifiability, consistency, and benchmarks. Each introduces assumptions about the data-generating process that are often untestable. We argue that there are no shared standards for comparing, prioritizing, or even discussing those assumptions. We support this claim by showing that (i) identifiability assumptions are often preferred by convention rather than by principle, (ii) uniform consistency requires further untestable assumptions, while also the weaker pointwise consistency has arguably narrowed algorithm design, and (iii) benchmarks are not neutral tests but carry an epistemic role of their own: they quietly fix which regimes matter, which guarantees come into play, and what counts as success. In doing so, they shape what algorithm comparisons are taken to show and thus perform some of the same epistemic work as theory, but less transparently. In this position paper we argue for a more transparent, integrated practice: Make the epistemic standards for discussing identifiability and consistency assumptions explicit, especially their failure modes and scope conditions; design benchmarks to operationalize those standards, by stress-testing and ablating both. Then, theoretical and empirical results jointly clarify when, why, and how algorithms may work rather than accumulate as disconnected advances.

Causal discovery is the task of leveraging observational data to uncover causal relationships between variables. Recent work has extended these methods to operate over clusters of variables to improve scalability in high-dimensions and enable reasoning over higher-level entities. These approaches have been limited by strong assumptions including causal sufficiency. In this work, we introduce an approach for causal discovery over clusters in non-Markovian systems. First, we extend theory of graphical models in a knowledge-based context, to motivate introduction of a novel graphical equivalence class that can accommodate unobserved confounding. Then, we present a sound algorithm for causal discovery of learnable relationships between clusters of variables.

Foundation Models (LLMs/VLMs) exhibit strong semantic reasoning capabilities but remain challenged by situated planning in unstructured physical environments. A key limitation is the Semantic-Geometric Gap: while models interpret linguistic and visual intent, they lack explicit grounding in continuous spatial structures, yielding physically infeasible or unsafe plans. We propose Causal-Geo, a neuro-symbolic framework that bridges this gap by integrating LLM reasoning with rigorous geometric planning. Rather than treating physical constraints heuristically, we model the environment as a continuous metric tensor field. LLMs translate high-level semantic intents into potential functions that dynamically warp this manifold via conformal scaling. This casts intent-driven planning as a geodesic optimization problem, solved efficiently via discrete graph search. Experiments across diverse domains—from a macro-scale $12.2 \text{ km}^2$ wetland digital twin to micro-scale indoor robotic navigation—demonstrate that Causal-Geo enables robust zero-shot planning. It significantly outperforms reinforcement learning in sparse-reward settings and prevents the ``physical hallucinations'' common in representative hierarchical LLM planners. Crucially, Causal-Geo acts as a rigorous physical gatekeeper, guaranteeing invariant safety against foundation model variance and demonstrating graceful degradation under contradictory prompts. These results establish continuous geometric grounding as a principled, domain-agnostic pathway toward physically consistent situated agent planning.

Discovering causal structures in multivariate time series (MTS) is critical in domains such as finance and neuroscience, yet remains difficult in practice due to three intertwined obstacles. First, MTS are routinely corrupted by missing values, and naive impute then discover pipelines can distort the underlying data distribution, leading to false causal graphs. Second, recovering instantaneous links is already hard on complete data, and missingness further entangles imputation with graph learning. Third, real world data often exhibits heteroscedastic noise, whose variance depends on both instantaneous and lagged causes, violating the assumptions of standard causal discovery algorithms. Existing methods fail to address these challenges concurrently, typically assuming complete data, homoscedastic noise, or purely time lagged relationships. To address these gaps, we introduce Cheesefill, an optimal transport framework for causal structure learning under missingness. Cheesefill parameterizes causal mechanisms with conditional normalizing flows to capture heteroscedastic noise, and jointly learns a stochastic correction map that refines naive imputations into trajectories consistent with the inferred mechanisms. Theoretically, we show that optimizing over stochastic correction maps is equivalent to a conditional kernel reformulation of the Kantorovich problem; the neural parameterization used by Cheesefill then yields a tractable objective. Extensive experiments on synthetic and real data show that Cheesefill consistently outperforms existing baselines across different missingness mechanisms.

As state-of-the-art neural networks are deployed on reasoning and algorithmic tasks, exactness guarantees become increasingly important. However, high average-case accuracy can still mask inconsistent behaviors. This motivates exact certification, which asks for the smallest set of labeled examples needed to certify that a learned hypothesis equals the target. We show that while some hypotheses are easy to certify, even minimal overparametrization can make certification exponentially hard across several hypothesis classes. For threshold circuits of depth $\ge 2$, adding a single extra gate can force certificate sizes exponential in the input dimension. We show an analogous hardness result for log-precision Transformers with only constant architectural overhead. We also characterize approximate certification, showing that allowing only polynomially many mistakes still requires exponentially large certificates, whereas constant relative-error guarantees can hide exponentially many failures. Empirically, we study certification for circuits and Transformers trained to recognize binary addition and find that imperfect models can evade detection unless the certificate is exponentially large.

LLM-based agents increasingly rely on external tools to accomplish complex tasks, yet the security of tool-calling pipelines remains poorly understood at the chain level. Prior attacks target individual tool invocations through prompt injection or metadata manipulation, but compromising a single step in a multi-step workflow is conspicuous and rarely sufficient for complex adversarial objectives. In this work, we uncover a more insidious yet realistic attack surface, tool-chain hijacking, in which an adversary constructs a coherent sequence of tools that collectively replace the agent's intended execution trace while still completing the user's task correctly, rendering the hijack invisible to both the agent and the user. To operationalize this threat, we propose ChainForge, an execution-grounded framework that mines agent execution logs to synthesize replacement chains through rollout-based optimization, then embeds adversarial payloads into the chain's code via multi-criteria iterative refinement. To systematically evaluate chain-level threats, we further construct ChainBench, a benchmark of 97 tasks across 4 domains. Experiments on four frontier LLMs show that ChainForge achieves a trace hijack rate of up to 98.54%, maintains task utility above 80.41%, transfers across models with over 84.95% success, and evades all evaluated defenses and code-safety scanners at substantially higher rates than single-tool baselines, exposing a critical blind spot in current agent security.

Instantaneous Quantum Polynomial Quantum Circuit Born Machines (IQP-QCBMs) have been proposed as quantum generative models that combine a classically tractable training objective—based on the maximum mean discrepancy (MMD)—with a potential quantum advantage motivated by sampling-complexity arguments. While recent works have explored this model across various application domains, fundamental questions remain: does the model suffer from exponentially vanishing loss gradients, known as the barren plateau problem—a pervasive obstacle in quantum machine learning—and how do regimes of trainability relate to regimes of possible quantum advantage? Here, we address both questions analytically. To study trainability, we derive closed-form expressions for the variances of the partial derivatives of the MMD loss function and establish general upper and lower bounds. We explicitly characterize how trainability depends on the generator set and the spectrum of the chosen kernel, identifying regimes in which low-weight kernels avoid exponential gradient suppression under structured topologies. Regarding potential quantum advantage, we reformulate the anti-concentration property in terms of the same generator-set quantities that govern trainability, enabling a unified analysis. We show that sparse IQP architectures can produce classically intractable output distributions while simultaneously remaining trainable, at least at lower-weight frequencies. Our analytical results corroborate the numerical observations of prior work and provides principled guidelines for designing scalable IQP-QCBM architectures.

Entities in real-world systems often evolve over time: users shift between information consumption patterns, researchers migrate between fields, and firms transition between financial risk states. When such systems are modeled as dynamic graphs, these transitions correspond to nodes whose class labels change between consecutive snapshots, which we call shifters. Correctly classifying shifters is often more consequential than classifying stable nodes with persistent labels, as detecting such transitions enables timely intervention in high-stakes settings. Yet, we identify a systematic failure mode across all major dynamic graph neural network architectures: shifter performance consistently lags behind stable nodes, with errors concentrated on the old label. Our structural analysis traces this failure to embedding inertia: neighborhood aggregation keeps shifter representations anchored to the old-class centroid, and this effect is strongest precisely where it is hardest to correct, namely for shifters whose neighborhoods remain aligned with the old class. Guided by this finding, we propose CHASE, a model-agnostic CHange-Aware framework for Shifting nodEs that wraps any dynamic GNN with targeted components for detecting label shifts and overriding stale neighborhood signals. CHASE consistently improves shifter performance across all tested models and datasets (up to 114.9% and 202.5% improvement on accuracy and F1 score, respectively) while preserving stable-node performance. We additionally contribute three dynamic graph benchmarks with naturally shifting labels, filling a gap in existing resources. The source code is available at https://anonymous.4open.science/r/CHASE-40CF/.

Human pose estimation has advanced rapidly driven by large-scale adult-centric models. However, children remain significantly underrepresented due to data scarcity, distinct body morphology, and unique motion dynamics. We introduce ChildPose, a modular video foundation model designed for child-centric pose estimation. ChildPose transforms pretrained pose encoders into a streaming pediatric model via memory-augmented temporal attention and age-aware co-training. By maintaining a memory bank of adjacent frame representations, the model achieves robust localization under motion blur, occlusion, and atypical poses, while an auxiliary age-prediction objective enforces representations that capture developmental morphology. We curate a diverse, video-based pediatric dataset spanning various age groups and capture conditions, evaluating out-of-distribution generalizability on entirely unseen subjects. Across multiple 2D baselines, ChildPose consistently outperforms both off-the-shelf models and child-data fine-tuned variants, while ensuring stable sequence-level predictions. Our results underscore that effective pediatric pose estimation necessitates child-specific modeling beyond standard adult-centric estimators.

Hallucination detection is critical for deploying large language models (LLMs) in real-world applications. Due to strong empirical performance, internal representation–based methods have emerged as the prevailing direction for detecting hallucinations, yet they remain largely model-specific and often fail to generalize to unseen LLMs. In this paper, we study an important yet underexplored problem, termed cross-model hallucination detection (CMHD), which aims to train hallucination detectors on source LLMs while ensuring performance on unseen target LLMs. The core challenge of CMHD lies in cross-model representation heterogeneity: hidden states from different LLMs exhibit severe semantic inconsistency, which limits the transferability of classical representation-based detectors. Through analysis and empirical validation, we show that relational structures over layers, tokens, and features capture transferable detection signals despite semantic inconsistency. Based on this observation, we propose cross-model hallucination detection via heterogeneity-oriented relational discrimination (CHORD), a relational graph discrimination framework that represents hidden states as joint relational graphs over layers, tokens, and features. It feeds these graphs into relation-aware attention to obtain structure-aware features, and uses meta-learning to shape these features toward transferable detection signals. Extensive experiments show that CHORD outperforms representative detection methods in cross-model generalization.

State estimation is a critical task in scientific, engineering and control applications. Since the reliability of reconstructions can depend on the number and position of sensors, Optimal sensor placement (OSP) is essential in scenarios where measurements are sparse and expensive. However, classical OSP approaches rely on Gaussian assumptions and are consequently unable to account for complex distributions encountered in many real-world systems. Generative-model-based reconstruction using sensor guided diffusion posterior sampling (DPS) has emerged as a promising technique for reconstructing states from highly complex distributions. Existing approaches to sensor selection either choose an unrealistically large number of sensors or employ strategies that emulate classical OSP methods. This results in a mismatch, wherein new models are paired with classical OSP tools, and motivates the need for fundamentally new ideas towards OSP that match the recent advances made in powerful recovery models. In this work, we introduce a distribution-free sensor placement framework based on the Christoffel function. Our main theoretical contributions introduce a mathematical formulation of optimal sampling and recovery guarantees for posterior sampling with arbitrary sensors and signal distributions. We use these to derive a new OSP strategy with non-asymptotic bounds on the number of sensors needed for recovery. Building on this, we develop \textbf{Christoffel-DPS}, with both offline and online variants, that implements nonparametric realizations of Christoffel sampling for generative models. As we show, Christoffel-DPS outperforms Gaussian OSP baselines and existing generative-model-based placement methods, validating that distribution-free sensing is both theoretically principled and practically superior. The framework is model agnostic, and we demonstrate its application to a range of unconditional DPS and flow matching models on structurally non-Gaussian benchmarks, showing the efficacy of Christoffel-DPS in low sensor budget regimes.

We present CivBench, an open-source benchmark for evaluating language model agents in long-horizon, tool-mediated environments through the Model Context Protocol (MCP). A single episode spans 300+ turns and produces thousands of tool calls over a large action space, requiring sustained planning, state monitoring, and execution under partial observability. The environment exposes 76 MCP tools and a narration layer that converts visual game state into structured text. We use CivBench to characterise agent behaviour across four model families in 23 admissible runs. The sample is a pilot, not a model ranking: aggregate outcomes do not reliably discriminate models at this scale. Instead, we introduce two interface-level metrics that the environment makes measurable: Proactive Monitoring Rate (PMR), capturing whether agents actively query latent strategic state, and RAG@10, capturing whether commitments stated in structured planning reflections are executed within ten subsequent turns. Across runs we observe two consistent patterns under a shared playbook protocol. Agents under-monitor strategically relevant state that is available but requires explicit querying: despite playbook guidance to query victory progress every 20 turns, agents do so only every 30–75 turns, and in 7 of 20 detectable defeats they failed to query within the 20-turn warning window before game end. Agents also frequently fail to execute near-term commitments stated in their own planning reflections (RAG@10 between 48% and 66% across models). Both patterns arise despite tool access and explicit guidance, and we interpret them as deviations under instruction rather than absences of capability. We release the environment, scenarios, logs, metrics, and analysis pipeline at https://anonymous.4open.science/r/civbench/README.md.

Pipeline-parallel distributed optimization is essential for large-scale machine learning but is challenged by significant communication overhead from transmitting high-dimensional activations and gradients between workers. Existing approaches often depend on impractical unbiased gradient assumptions or incur sample-size memory overhead. This paper introduces Clapping, a Communication compression algorithm with LAzy samPling for Pipeline-parallel learnING. Clapping adopts a lazy sampling strategy that reuses data samples across steps, breaking sample-wise memory barrier and supporting convergence in few-epoch or online regimes. Clapping comprises two variants including Clapping-FC and Clapping-FU, both of which achieve convergence without unbiased assumption for compressed gradient, effectively addressing compression error propagation in multi-worker settings. Numerical experiments validate the performance of Clapping across different tasks.

Large-scale pre-trained vision models provide powerful representations for class-incremental learning, yet continuously adapting them to new classes without historical samples, task identifiers, or unrestricted parameter growth remains a fundamental challenge. Existing LoRA-based approaches typically follow either a one-adapter-per-task expansion paradigm or a fixed parameter-sharing strategy. The former leads to linear parameter growth as the task sequence expands, while the latter often induces severe cross-task interference. To address these limitations, we propose ${\bf E}^3$ (Elastic Ensemble of Experts), a parameter-efficient framework for class-incremental learning. ${\bf E}^3$ treats LoRA modules as elastic experts whose task capacity is dynamically scheduled according to representational saturation. Within each expert, we further introduce Group-Aware Parameter Partitioning, which allocates LoRA parameters into disjoint task-specific subspaces using magnitude-based importance estimation, thereby mitigating parameter overwriting without additional gradient-sensitivity computation. Moreover, ${\bf E}^3$ incorporates hierarchical moment regularization and orthogonal gradient projection to constrain classifier drift and prevent new-task updates from disrupting historical feature subspaces. Extensive experiments on standard class-incremental learning benchmarks demonstrate that E³ achieves state-of-the-art or competitive performance, especially in long-horizon task sequences, while maintaining a favorable balance between stability, plasticity, and parameter efficiency.

The contemporary LLM-agent stack can automate substantive criminal activity at scale by composing capabilities that are already present, and the relevant engineering and governance controls need to be in place before that composition appears in production. The argument rests on three empirical legs that are usually studied separately but jointly create a closed loop of accidental or adversarial criminal retail automation. (i) Autonomous LLM agents fail catastrophically at well-studied multi-agent coordination tasks, with bankruptcy as the modal outcome rather than a tail event (ii) The rate at which agents elicit clearly harmful tools is not fixed but moves with common contextual manipulations. (iii) Autonomous browser control has become general enough that frontier agents reliably navigate, populate, and complete checkout flows on simulated illicit-product storefronts when light obfuscation is applied. Combining the three legs requires no new model capability: in our experiments, LLM agents autonomously completed simulated dark-web purchases of drugs, shotguns, and hitman services end-to-end. We propose three interventions: mandatory pre-deployment safety testing, hard human-confirmation gates at financially material steps, and verifiable agent ID infrastructure.

Multimodal Large Language Models (MLLMs) are prone to hallucination as their generation preferences are insufficiently calibrated to visual evidence, causing them to fall back on linguistic priors, rather than faithful grounding. In this work, we start from an empirical observation: when query-relevant visual evidence is explicitly strengthened using the model’s own attention, generation becomes more accurate, suggesting that many failures do not arise solely from missing perception, but from an insufficient tendency to trust the evidence the model has already attended to. Motivated by this finding, we propose Oriented Pickup Preference Optimization (\texttt{OPPO}), an evidence-aware alignment objective that learns preferences over the strength of visual evidence, rather than only response quality. Concretely, \texttt{OPPO} contrasts the same faithful response under stronger, anchored, weaker-evidence views, turning naive visual preference into ordered visual-evidence alignment. We further combine this objective with fine-grained span-level and token-level regularization to stabilize the training. Besides, we provide a theoretical analysis showing that ordered evidence margins induce a positive lower bound on local visual sensitivity. Extensive evaluations across hallucination and general-purpose benchmarks demonstrate that \texttt{OPPO} consistently outperforms baseline methods.

We study 3D visual grounding in 3D Gaussian Splatting (3DGS), where a referring expression should identify a target object as both a rendered 2D mask and a set of explicit 3D Gaussians. Prior methods typically extract 3D targets by applying heuristic thresholding or top-ratio filtering to dense primitive-level relevance scores. However, these scores often vary substantially across scenes and prompts, making such heuristics sensitive and leading to unstable targeting. ClusterSplat addresses this by reframing 3D target selection as cluster-level selection over scene-adaptive candidates. The method first learns instance-aware Gaussian features, forms feature-aware seed voxels, and merges neighboring seed voxels when a description-length criterion decreases, producing a scene-specific set of cluster candidates. Given a referring expression, a query-conditioned cluster scorer ranks the cluster candidates with a lightweight MLP, and the selected cluster directly defines the explicit 3D target while its cluster score is rendered for the 2D mask. On ScanRefer, ClusterSplat achieves the best rendered 2D segmentation and explicit 3D target metrics among the compared 3DGS baselines. It also obtains the best average Ref-LERF scores, supporting the same trend under fine-grained referring descriptions. On ScanNet v2 class name queries, ClusterSplat achieves the strongest rendered 2D results and the highest 3D matching accuracy, while maintaining competitive 3D IoU.

Molecular graph generation typically involves a trade-off between two paradigms: diffusion models capture global structure well but require hundreds of denoising steps, while autoregressive (AR) models sample in a single pass yet generate atoms in a flat canonical order with no semantic hierarchy, leaving long-range topology to chance. We introduce the **Hierarchical Graph VQ-Transformer (H-GVT)**, which removes this dichotomy by performing coarse-to-fine AR generation over hierarchical discrete codes. A **Spectrally-Regularized Multi-Scale VQ-VAE** first compresses each molecule into a sequence of discrete tokens at multiple scales, where coarse tokens provide regularized low-resolution structural context and fine tokens encode atom-level details. The compression combines Adaptive Min-Cut Pooling with a novel **Spectral Topology Consistency Loss** that aligns low-frequency normalized Laplacian spectra across scales, adding only 2.8% training overhead. A standard decoder-only Transformer with level embeddings then generates these tokens from coarse to fine, allowing fine-grained atom tokens to condition on coarser structural context. This requires no specialized tree, motif, or scale-causal masking. On QM9 and ZINC250k, H-GVT achieves the best NSPDK among compared baselines ($2\times10^{-4}$ on QM9 and $1\times10^{-4}$ on ZINC250k), while sampling 10K molecules $68\times$ faster than DiGress. On MOSES, H-GVT achieves $4.8\times$ better FCD than a Novelty-thresholded GVT baseline under the same thresholded protocol (0.19 vs. 0.92; 86.4% vs. 80.5% Novelty). Crucially, ablations reveal that the coarse-to-fine ordering itself, not merely multi-scale tokenization, is what enables global structural planning: randomizing the order while keeping the same tokens, codebook, and level embeddings degrades NSPDK by $29\times$.

Fine-tuning has become a central mechanism for adapting large language models (LLMs) to downstream tasks, yet most existing methods follow a one-to-one paradigm: a single pretrained model is optimized into a single improved policy. This paradigm optimizes average single-policy performance, leaving no mechanism to distinguish redundant successes from genuinely complementary contributions. In this paper, we propose Co-Evolve, a one-to-many LLM fine-tuning framework that transforms a single base model into a cooperative population of persistent, specialized descendants and explicitly optimizes their complementarity. To reduce redundancy among descendants, we introduce a marginal-coverage objective that rewards each model for paying attention to examples under-covered by its peers. We instantiate this objective with evolution strategies, enabling derivative-free population-level optimization. We further introduce a seed-replay checkpoint mechanism that reduces the disk storage required to maintain multiple descendants by avoiding storing multiple full checkpoints during deployment. Experiments across multiple domains show that Co-Evolve consistently improves ensemble-level performance over strong fine-tuning and ensemble baselines. Our results suggest that optimizing for complementary model-level specialization provides a scalable alternative to single-policy fine-tuning. Our code can be found at https://anonymous.4open.science/r/Co-Evolution-Core-6604.

Robots operating in semantic environments often need to satisfy linear temporal logic (LTL) tasks before the semantic map is fully known. Many practical semantic-LTL planners maintain semantic beliefs but execute a single task policy induced by a point semantic interpretation, such as a maximum-a-posteriori (MAP) semantic map, possibly supported by risk checks or active perception. This premature commitment can be brittle when distant semantic observations are weak or correlated. We propose Coherence-Aware Transition-Intent Fusion, an online planner that preserves automaton-relevant semantic uncertainty using a finite top-$K$ intent abstraction. Given a conservative current automaton estimate, the planner propagates plausible semantic assignments through the automaton, converts feasible successor transitions into weighted reach-avoid intents, and fuses their risk-gated Dijkstra progress scores. When competing intents create incoherent motion, a KL-regularized calibration step shifts weight toward intents consistent with recent physical progress. Experiments on random-obstacle and room-like semantic maps show improved satisfaction over single-intent and raw-fusion variants, and over a TOAPP* MAP-plus-active-perception baseline under the same static-view sensor. Ablations attribute the gain primarily to intent-level calibration.

Recent Thinking with Video approaches use Video Generation Models (VGMs) for visual reasoning by producing temporally coherent Chain-of-Frames as reasoning artifacts. Even strong VGMs, however, exhibit two recurring failure modes on goal-directed tasks: long-horizon drift on multi-step tasks and mid-clip simulation errors that compound. Both stem from the absence of explicit reasoning built upon the VGM's short-horizon visual prior, a role naturally filled by Vision-Language Models (VLMs), but where to place the VLM is non-trivial: upfront plans commit before any frame is generated and post-hoc critiques over whole videos intervene too late. We propose VLM-VGM Collaborative Video Reasoning (CollabVR), a closed-loop framework that couples the VLM with the VGM at step-level granularity: the VLM plans the immediate next action, inspects the clip the VGM generates, and routes test-time compute across qualitatively distinct recovery strategies (re-generation, action splitting) matched to the diagnosed failure. On Gen-ViRe and VBVR-Bench, CollabVR improves both open-source and closed-source VGMs over single-inference, Pass@k, and prior test-time scaling baselines at matched compute, with the largest gains on the hardest tasks. It also yields further improvements on top of a reasoning-fine-tuned VGM, indicating that step-level VLM supervision is orthogonal to and stackable with reasoning-oriented fine-tuning. We provide video samples and additional qualitative results at our project page: https://collab-vr.github.io.

Due to the potential for exploratory reasoning of Latent Visual Reasoning, recent works tend to enable MLLMs (Multimodal Large Language Models) to perform visual reasoning by propagating continuous hidden states instead of decoding intermediate steps into discrete tokens. However, existing works typically rely on hard alignment objectives to force latent representations to match predefined visual features, thereby severely limiting the exploratory of latent reasoning process. To address this problem, we propose CoLVR (Contrastive Optimization for Latent Visual Reasoning). To obtain a more exploratory visual reasoning, CoLVR introduces a latent contrastive training framework. Firstly, CoLVR learns diverse and exploratory representations with a latent contrastive objective guided by angle-based perturbation, which expands the semantic latent space and avoids over-constrained embedding. Then, CoLVR employs a latent trajectory contrastive reward for RL (Reinforcement Learning) post-training to enable fine-grained optimization of latent visual reasoning process and thus fostering diverse reasoning behaviors. Experiments demonstrate that CoLVR significantly enhances the exploratory capability of latent representations, achieving average improvements of 5.83% on VSP and 8.00% on Jigsaw, while also outperforming existing latent models on out of domain benchmarks, with a 3.40% gain on MMStar.

Post-hoc unauthorized-training data detection for large language models (LLMs) typically assumes a query-with-originals regime: rights holders query a target LLM with raw proprietary data and assess whether the model assigns them stronger memorization-based detection signals, e.g., higher confidence or lower loss, than held-out non-training reference texts. We show that this regime becomes brittle under data laundering, where the target LLM is trained on semantics-preserving but stylistically or structurally transformed surrogates of proprietary data to obfuscate provenance. Since training-time exposure occurs in the laundered form, memorization signals may no longer appear on the originals, collapsing the candidate-reference signal separation that standard detectors rely on. We counter this threat by studying laundering-aware detection with raw proprietary data, a held-out reference corpus, and query access to the target LLM, while the laundering transformation is undisclosed. Since exact recovery of the laundered corpus is infeasible, we infer a detection-useful synthesis process via an auxiliary LLM that maps originals into training-like queries. To make this search tractable, we introduce Synthesis Data Reversion (SDR), which constrains the unbounded space of natural-language transformations through a goal-details decomposition: a high-level transformation goal, e.g., "lyrical rewriting", and fine-grained details, e.g., "with vivid imagery". SDR identifies the most likely goal and iteratively refines details so synthesized queries elicit stronger target-model detection signals. Evaluated on the MIMIR benchmark against diverse laundering practices and target LLM families (Pythia, Llama2, and Falcon), SDR consistently restores detection signals, offering a practical auditing tool against data laundering.

Electronic health records (EHRs) contain rich health information for uncovering latent patient representations and clinically meaningful subgroups. Learning from EHR data across institutions is valuable because representations learned at a single site may reflect local data patterns and lack generalizability, yet pooling patient-level records across institutions is often impractical because of privacy constraints and communication costs. We propose a communication-efficient, non-iterative federated spectral method for learning shared latent representations from multi-institutional EHRs, in which each institution shares aggregate marginal frequencies and a low-rank spectral summary while keeping patient-level records local. A central server aggregates these summaries with the pooled marginal frequencies to recover both the shared low-rank structure and the canonical row-normalization direction required for interpretable representation recovery. We prove row-wise recovery guarantees and error bounds, with a leading federated variance term matching a corresponding lower bound under explicit regularity conditions. Simulations show that our method closely matches pooled analysis in estimating latent representations, improves as more sites contribute summaries, and outperforms single-site and existing federated benchmarks across a range of data settings. Applied to real-world multi-facility EHR data for stroke patients, the method learns interpretable patient representations that highlight clinically distinct subgroups and could help inform subgroup-specific care.

Parameter-efficient fine-tuning methods, such as LoRA, offer a practical way to adapt large vision and language models to client tasks. However, this becomes particularly challenging under task-level heterogeneity in federated deployments. In this regime, personalization requires balancing general knowledge with personalized knowledge, yet existing approaches largely rely on heuristic mixing rules and lack theoretical justification. Moreover, prior model merging approaches are also computation and communication intensive, making the process inefficient in federated settings. In this work, we propose Potara, a principled framework for federated personalization that constructs a personalized model for each client by merging two complementary models: (i) a federated model capturing general knowledge, and (ii) a local model capturing personalized knowledge. Through the construct of linear mode connectivity, we show that the expected task loss admits a variance trace upper bound, whose minimization yields closed-form optimal mixing weights that guarantee a tighter bound for the merged model than for either the federated or local model alone. Experiments on vision and language benchmarks show that Potara consistently improves personalization while reducing communication, leading to a strong performance-communication trade-off.

Chunked prefill has become a widely adopted serving strategy for long-context large language models, but efficient attention computation in this regime remains challenging. Existing sparse attention methods are primarily designed for one-shot prefill and do not translate efficiently to chunked prefill: block-sparse kernels lose efficiency when the query length is limited by the chunk size, while fine-grained pattern search becomes costly when repeated over the accumulated KV cache at every chunk. QUOKA, a recent method that directly targets chunked prefill, avoids sparse-kernel overhead but relies on query-subsampled, token-level KV selection, which can miss query-specific KV entries and introduce explicit KV-copy overhead. To address these limitations, we propose CompactAttention, a chunked-prefill attention mechanism based on Block-Union KV Selection. CompactAttention treats 2D block-sparse masks as KV-selection signals rather than direct sparse-kernel execution plans, and converts them into GQA-aware per-group KV block tables through Q-block union and intra-group union. This construction produces the minimal block tables that preserve all KV blocks selected by the input masks under paged execution constraints, enabling selected KV blocks to be accessed in place without explicit KV compaction. On LLaMA-3.1-8B-Instruct, CompactAttention maintains accuracy close to dense attention on the RULER benchmark while delivering up to 2.72$\times$ attention speedup at 128K context length under chunked prefill.

SO(3) equivariant graph neural networks have become the dominant paradigm for atomistic foundation models, achieving high accuracy and data efficiency by building rotational symmetry directly into the architecture. Yet the computational cost of their higher-order tensor operations creates a tough trade-off between model accuracy and inference efficiency. In this paper, we propose a structural pruning method for SO(3) equivariant atomistic foundation models to bridge this accuracy-efficiency gap. The pruning is applied along the channel and order dimensions, with each irreducible representation kept or removed as a complete block, thereby retaining SO(3) equivariance. Starting from a large checkpoint, the pruned model substantially reduces the inference cost while retaining higher accuracy than an independently trained small model. The pruned MACE-MP model outperforms the official from-scratch trained small model on 7 of 9 metrics on the Matbench Discovery leaderboard. In terms of efficiency, compressed MACE-MP and MACE-OFF models contain 1.5$\times$ to 4$\times$ fewer parameters and require 2.5$\times$ to 4$\times$ less pre-training compute than training a small model from scratch. For downstream applications, fine-tuning the pruned model reduces energy and force errors by 70.1\% and 34.4\% compared to training task-specific models from scratch across eight representative downstream datasets. We demonstrate that the method generalizes to other SO(3) equivariant architectures (SevenNet, eSCN) and can be combined with quantization and knowledge distillation for further gains.

A unified cross-modal person re-identification targets robust cross-modal retrieval with queries from diverse modalities like visible, infrared, and text.Such framework requires multi-modal data and centralized training, raising data costs and privacy risks. Hence, we propose a modality-heterogeneous federated ReID task to learn a unified cross-modal framework from clients with different cross-modal data.However, it faces two challenges.First, varying inter-modality complexity across tasks leads to different retrieval patterns and parameter space. Second, different intra-modality complexity results in imbalanced contributions when merging LoRA modules.Existing federated learning methods use uniform merging and transmit full models, which hinders cross-modal knowledge integration and increases communication overhead. To address these issues, we adopt Low-Rank Adaptation (LoRA), a lightweight module for fine-tuning large models through low-rank updates, enabling clients to send compact parameters.On top of LoRA, we develop MIRAGE (Modality-heterogeneous Intelligent federated ReID Aggregation with Global Efficiency), adapting aggregation to inter- and intra-modality complexity.When merging visible LoRA, MIRAGE selects key ranks to protect retrieval patterns while retaining shared ranks for common ability.As for infrared and text LoRA, it reduces conflicts with an adaptive pruning method. In both strategies, task complexity is incorporated into aggregation to balance contributions. Experiments on cross-modal retrieval tasks demonstrate the superiority of MIRAGE.The code and dataset will be publicly released.

Out-of-Distribution (OOD) detection requires sensitivity to subtle shifts without overreacting to natural In-Distribution (ID) diversity. However, from the viewpoint of detection granularity, global representation inevitably suppresses local OOD cues, while patch-based methods are unstable due to entangled spurious-correlation and noise. And neither of them is effective in detecting compositional OODs composed of valid ID components. Inspired by recognition-by-components theory, we present a post-hoc Component-Based OOD Detection (CoOD) framework that addresses the existing limitations by decomposing inputs into functional components. To instantiate CoOD, we derive Component Shift Score (CSS) to detect local appearance shifts, and Compositional Consistency Score (CCS) to identify cross-component compositional inconsistencies. Empirically, CoOD achieves consistent improvements on both coarse- and fine-grained OOD detection.

Reconstructing the motion of objects from videos is a key component for embodied AI and robot manipulation. While diverse approaches to object pose tracking have been studied, they rely heavily on strong external priors, such as depth data or 3D templates, and remain highly vulnerable to severe occlusions by hand grasps despite the use of explicit masks. In this work, we present ComPose, a 6DoF object tracking framework designed for hand-aware object pose estimation from RGB video. Rather than treating the hand purely as an occluder, our method harmonizes hand motions as a complementary cue for object tracking. In detail, we recover a variety of object motions over time by combining object- and hand-derived cues from foundation models within a unified tracking pipeline. Here, ComPose adaptively selects informative hand joints, blends object- and hand-based rotation cues, and refines translation from the visible geometric evidence using a learned correction. We further enforce the temporal consistency over both rotation and translation, yielding stable 3D object trajectories over time without any external smoothing. Extensive experiments show that our method is accurate, efficient, and robust under severe hand occlusion and geometric ambiguity. In addition, the resulting trajectories can also effectively transfer to downstream robot manipulation by enabling robots to reconstruct human actions from online videos.

Learning from multiple imperfect oracles is a practical way to improve reinforcement learning (RL) under limited interaction. However, most contemporary robust multi-oracle methods are on-policy and equire frequent fresh rollouts, which limits sample efficiency in domains where data collection is expensive. We propose Oracle-Aggregated Policy Improvement (OPI), an off-policy framework for policy improvement with multiple suboptimal oracles. OPI augments the oracle set with the learner, constructs an oracle-guided behavior policy through candidate-action pooling and source-aware critic scoring, and trains the learner using a generalized advantage-weighted behavior cloning objective combined with deterministic policy gradients. The method is motivated by a tractable proxy for the max-advantage objective and a policy-improvement relative to a strong oracle baseline. We evaluate OPI on diverse tasks across MetaWorld, the DeepMind Control Suite, and customized Box2D environments with heterogeneous oracles. Across sparse-reward manipulation, dense-reward locomotion, and customized control settings, OPI is strongest in sparse-reward and complementary-oracle regimes and remains competitive with representative baselines in dense-reward environments. Additional analyses show that oracle composition is most beneficial when oracle skills are complementary and that estimating oracle-specific action values is particularly helpful in sparse-reward regimes

Large-scale verifiable prompts underpin the success of Reinforcement Learning with Verifiable Rewards (RLVR), but they contain many uninformative examples and are costly to expand further. Recent studies focus on better exploiting limited training data by prioritizing hard prompts whose rollout pass rate is 0. However, easy prompts with a pass rate of 1 also become increasingly prevalent as training progresses, thereby reducing the effective data size. To mitigate this, we propose Composition-RL, a simple yet useful approach for better utilizing limited verifiable prompts targeting pass-rate-1 prompts. More specifically, Composition-RL automatically composes multiple problems into a new verifiable question and uses these compositional prompts for RL training. Extensive experiments across model sizes from 4B to 30B show that Composition-RL consistently improves reasoning capability over RL trained on the original dataset. Performance can be further boosted with a curriculum variant of Composition-RL that gradually increases compositional depth over training. Additionally, Composition-RL enables more effective cross-domain RL by composing prompts drawn from different domains. All our codes, compositional datasets, and models will be released to facilitate future research.

This position paper argues that concept-based mechanistic interpretability lacks a concrete evaluation paradigm and that the field must converge on one before continuing to scale methods whose effectiveness remains unverified. Sparse Dictionary Learning (SDL), encompassing sparse autoencoders, transcoders, crosscoders, and their variants, has become the dominant methodology for concept-based mechanistic interpretability, supported by a growing evaluation suite: reconstruction loss, automated interpretability scoring, human evaluation, feature absorption rates, sparse probing, and model editing benchmarks. Despite the significant empirical success SDL has achieved, we argue that none of these evaluation metrics directly measures what SDL is actually supposed to achieve: recovering the ground-truth features underlying observed representations. This insufficiency is structural, not incidental, and we support this perspective with both theoretical and empirical evidence. On the theoretical side, we build on the most recent formal analyses of SDL in mechanistic interpretability to show that the optimization landscape provably admits failure modes that existing metrics cannot detect. On the empirical side, we present three lines of evidence demonstrating that methods spanning a wide range of ground-truth recovery performance appear indistinguishable under current evaluation practice. Together, these results call for a new evaluation paradigm. We propose a set of principles that an ideal evaluation benchmark for concept-based mechanistic interpretability should satisfy, and argue for the community to develop and adopt more benchmarks grounded in these principles before continuing to scale methods whose effectiveness remains unverified.

Circuit extraction identifies model components that preserve a target behavior under ablation, yet it remains unclear which parts of the reported circuit—components, edges, or coarser summaries—remain stable across reasonable extraction choices. We study this question in a Lean-style, rule-generated tactic-prediction benchmark spanning atomic and compositional proof tasks. The proof-state rules and task structure are fixed by construction, and dense and weight-sparse variants of one transformer architecture are compared. In this controlled setting, we compare three ways to describe a circuit: a compact prediction-preserving circuit, a broader graph that keeps read, write, and routing components around that circuit, and a pruning graph whose size is set by a target post-ablation loss budget. We vary model sparsity, whether the query and key sides of attention are represented together or separately, the post-ablation loss budget for the pruning graph, and which supervised checkpoint initializes reinforcement learning (RL). When the same extraction method is used, dense and weight-sparse checkpoints overlap far more in the set of attention heads identified than in exact component-to-component edge lists. This head-level overlap is well above a random top-$k$ baseline; the exact-edge overlap is not consistently so. Tracking query and key sides separately, and varying the loss budget across the tested range, preserves the ordering of the RL initialization conditions when each graph is summarized by how many components it selects. Among these RL conditions, the largest gains on compositional tasks are accompanied by the largest fraction of compositional-task circuit nodes that lie outside the matched atomic-task circuits. These findings show that, even in this controlled setting, behavior preservation under ablation does not determine a unique circuit-level conclusion: the conclusion depends on which graph is reported, how it is extracted, and at which level of description the comparison is made.

A complete understanding of heterogeneous treatment effects requires fully capturing the conditional distribution of the counterfactual outcomes. To this end, we propose the Conditional Counterfactual Mean Embeddings (CCME), a framework that embeds conditional distributions of counterfactual outcomes into a reproducing kernel Hilbert space (RKHS). Under this framework, we develop a two-stage meta-estimator for CCME that accommodates any RKHS-valued regression in each stage. Based on this meta-estimator, we develop three practical CCME estimators: (1) Ridge Regression estimator, (2) Deep Feature estimator that parameterizes the feature map by a neural network, and (3) Neural-Kernel estimator that performs RKHS-valued regression, with the coefficients parameterized by a neural network. We provide finite-sample convergence rates for all estimators, establishing that they possess the double robustness property. Our experiments demonstrate that our estimators accurately recover distributional features including multimodal structure of conditional counterfactual distributions.

Retrieval-Augmented Generation (RAG) improves the factual accuracy of large language models (LLMs) by grounding responses in external evidence. However, when retrieved context conflicts with models' internal parametric knowledge, LLMs may still generate answers that contradict the provided evidence, posing a key challenge to contextual faithfulness and the reliability of RAG systems. Motivated by recent mechanistic findings on the distinct propagation and progressive accumulation of parametric and contextual signals in LLMs, we propose Conflict-Suppressed RAG (CSRAG), a simple, training-free, decoding-time framework for resolving knowledge conflicts. CSRAG biases generation toward retrieved evidence by suppressing tokens associated with parametric knowledge while boosting tokens from the context via two complementary logits processors. Experiments on six challenging faithfulness benchmarks demonstrate that CSRAG consistently achieves state-of-the-art or near state-of-the-art performance across multiple backbone LLMs, while remaining fully training-free and lightweight.

Deep-learning protein structure predictors achieve near-experimental accuracy on individual folds, yet their default inference samples concentrate around a single dominant conformation. We introduce ConforFlux, an inference-time procedure for Boltz-2 that couples $M$ structure-prediction trajectories through a pairwise C$\alpha$-RMSD repulsion gradient on the trunk's single and pair embeddings. Because the trunk conditions every block of the diffusion module, this update propagates to every subsequent denoising step. On four conformational-change categories, ConforFlux improves the per-state success rate over Default Boltz-2 by 3–17 percentage points while preserving Default-comparable physical quality. On twelve transporter pairs with at least one alternate-state reference released after the Boltz-2 cutoff, the $2$ Å success rate rises from $4/12$ to $9/12$. On the human dopamine transporter, ConforFlux additionally reaches the post-cutoff inward state, which $500$ Default samples never attain.

While diffusion and flow-based generative models have achieved impressive performance in high-fidelity image and video synthesis, this capability comes at the cost of substantial inference overhead during iterative sampling. Cache-based acceleration alleviates this cost by reusing intermediate computations across adjacent denoising or transport steps, typically relying on lightweight proxy discrepancies to decide when cached outputs should be refreshed. However, such proxy-discrepancy rules are inherently unreliable: the proxy is only a point estimate of the inaccessible oracle discrepancy and can misalign across prompts, timesteps, and generation dynamics. This mismatch induces asymmetric failures: proxy underestimation may cause unsafe reuse of stale cached computations and degrade generation quality, while proxy overestimation leads to overly conservative refreshes and reduced acceleration. To address these issues, this paper introduces Conformal Cache, dubbed CCache, a training-free conformal calibration framework for reliable cache-based generation acceleration. Our key insight is to formulate cache reuse as a one-sided risk-control problem and adaptively calibrate a timestep-wise upper correction for the signed proxy-to-oracle gap based on split conformal prediction. CCache is plug-and-play and can be integrated with different proxy-discrepancy caching baselines without structural modifications or parameter updates to the generative model. Extensive experiments on image and video generation demonstrate that \methodname improves generation quality and cache reliability while preserving favorable speed-quality trade-offs.

Gradient descent for matrix factorization exhibits an implicit bias toward approximately low-rank solutions, even in regimes where the iterates may grow unbounded. We study this behavior through a constrained factorization with diagonal scaling, separating bounded outer factors from explicit diagonal scale parameters. For positive semidefinite matrix recovery, we use the model $X \approx UDU^\top$, where $U$ is constrained to a Frobenius norm ball and $D$ is a nonnegative diagonal factor. Although this reparameterization preserves the stationary points of classical Burer-Monteiro factorization, its projected dynamics consistently recover truly (rather than approximately) low-rank solutions across a wide range of step sizes and initializations. Motivated by this behavior, we extend the construction to neural networks by inserting diagonal layers between Frobenius-norm-constrained outer layers. The resulting UDV models exhibit strong rank-revealing structure during training, with many diagonal components and associated columns collapsing toward zero and becoming naturally prunable. We exploit this structure through an SVD-based pruning procedure (UDVPruning) that identifies and removes redundant units during training. Across ViT and MLP-Mixer benchmarks, UDVPruning substantially reduces parameters and FLOPs while maintaining competitive test accuracy and often improving measured training time, inference latency, and energy consumption.

In this work, we develop proximal preconditioned gradient methods with a focus on spectral gradient methods providing a proximal extension to the Muon and Scion optimizers. We introduce a family of stochastic algorithms that can handle a wide variety of convex and nonconvex constraints and study its convergence under heavy-tailed noise, through a novel analysis tailored to the geometry of the proposed methods. We further propose a variance-reduced version, which achieves faster convergence under standard noise assumptions. Finally, we show that the polynomial iterations used in Muon are more accurately captured by a nonlinear preconditioner than by the ideal matrix sign, leading to a convergence analysis that more faithfully reflects practical implementations.

Real-world generative problems in biology, medical imaging, or robotics, rarely come with perfectly paired source and target distributions. While unbalanced Optimal Transport (uOT) provides a robust framework to bridge unpaired datasets via minimum-length paths in probability space, existing formulations based on the Wasserstein–Fisher–Rao (WFR) geometry introduce a structural bias toward high density regions, often triggering mode collapse. We introduce Contact unbalanced Optimal Transport (CuOT), a novel uOT formulation grounded in contact geometry that decouples density transport from mass growth to eliminate structural bias. This allows CuOT to adapt sampling steps to the local data structure, naturally reducing step sizes in high-density regions to prevent overshooting and ensuring stable convergence. The result is a stable, scalable path-length minimization solver that consistently outperforms WFR-based methods on biological dynamics reconstruction, unpaired image-to-image translation, and video generation.

Autoregressive image generation has recently emerged as a competitive visual generation paradigm, yet its generation order is often predefined or heuristic, such as next-patch prediction, random-order prediction, or one-step generation for specific scales. We observe that these strategies can cause models to overlook important structural dependencies in images, thereby limiting both structural robustness and reasoning capability. In this work, we propose a context-aware autoregressive generation framework, where the model dynamically determines the next token location based on global context and current local information. Rather than following a fixed or random order, our method allows the generation process to adapt to image structure and semantic dependencies. Extensive experiments show that context-aware generation significantly improves structural robustness across different image AR paradigms, including masked autoregressive and scale-wise autoregressive models. Beyond quantitative improvements, we further observe emerging reasoning properties: the model can better understand complex semantic instructions and generate images that more faithfully follow the given guidance. These findings suggest that generation order is a critical yet underexplored factor in image autoregressive modeling, and that context-aware decoding can unlock stronger reasoning-aware visual generation capabilities.

Multimodal models deployed in real-world settings often suffer from missing data modalities due to acquisition costs, privacy constraints, or sensor failures, leading to severe performance degradation. Existing approaches based on shared representations or expert routing struggle when modality-specific information is absent. A key challenge is that, when a modality is unobserved, the target representation is not uniquely identifiable, making naive latent prediction prone to degenerate or collapsed solutions. We propose Cross-modal Embedding Prediction and Alignment (CEPA), a framework that addresses missing modalities through task-driven latent representation imputation rather than raw input synthesis. CEPA employs masked representation learning with data-driven masking patterns and a context-conditional distribution alignment objective that stabilizes latent prediction and prevents representation collapse. We evaluate CEPA on the MIMIC multimodal benchmark spanning EHR time-series, chest X-rays, and clinical notes, across three clinical prediction tasks under both controlled (MCAR) and naturally occurring (MNAR) modality absence. CEPA consistently outperforms prior missing-modality methods, with ablations confirming the contribution of adaptive masking and context-conditional alignment.

Vision-language-action (VLA) models show promising knowledge accumulation ability from pretraining, yet continual learning in VLA remains challenging, especially for efficient adaptation. Existing continual imitation learning (CIL) methods often rely on additional parameters or external modules, limiting scalability for large VLA models. We propose Stellar VLA, a knowledge-driven CIL framework without increasing network parameters. Two progressively extended variants are designed: T-Stellar for flat task-centric modeling and TS-Stellar for hierarchical task–skill structure. Stellar VLA enables self-evolving knowledge learning by jointly optimizing task representations and a learned knowledge space. We propose a knowledge-guided expert routing mechanism conditioned on knowledge relation and Top-$K$ semantic embeddings, enabling task specialization without increasing model size. Experiments on the LIBERO benchmark show that Stellar VLAs achieve strong performance among both VLA and CIL baselines, using only 1\% data replay. Real-world evaluation on a dual-arm platform with distinct embodiment and scene configurations validates effective knowledge transfer. TS-Stellar excels in hierarchical manipulation, and visualizations reveal robust knowledge retention and task discovery. Project website is provided in the supplementary material.

The strong zero-shot and long-context capabilities of recent Large Language Models (LLMs) have enabled highly effective list-wise reranking systems. Attention-based rerankers leverage transformer attention patterns as retrieval signals, but not all attention heads contribute equally: many introduce noise and redundancy, limiting both reranking quality and efficiency. In this work, we introduce CoRe heads, a small subset of retrieval-specialized heads identified through a contrastive scoring objective that rewards attention to relevant documents while penalizing attention to distractors. This contrastive criterion isolates highly discriminative retrieval heads and yields a strong training-free list-wise reranker. Our analysis reveals that retrieval-relevant computation in transformer rerankers is highly sparse and structurally localized: fewer than 1% of attention heads account for most reranking performance across models and datasets, and these heads consistently concentrate in middle transformer layers. This localization enables aggressive layer pruning for efficient reranking: removing the final 50% of transformer layers preserves nearly identical reranking accuracy while substantially reducing inference latency and GPU memory usage.

Model merging integrates task-specific fine-tuned models into a single multi-task model, but often suffers from parameter interference caused by conflicting task-vector updates. Existing methods typically mitigate conflicts by pruning task vectors based on weight magnitude or random heuristics, treating Transformers as unstructured ``bags of parameters'' and overlooking their inherent modularity. In this paper, we propose \textbf{C}ontribution-\textbf{A}ware \textbf{S}tructured \textbf{S}parsity (CASS), a unified framework that reduces parameter interference by identifying and preserving task-specific components. At the core of CASS is a contribution-aware structured mask that identifies task-relevant attention heads and FFN neurons. We instantiate this mask in two settings: CASS-Merging, the primary post-hoc setting where masks serve as a plug-and-play denoising filter for existing merging operators, and CASS-Tuning, an extension for scenarios with fine-tuning access where masks constrain gradients to reduce structural overlap between task vectors. Our analysis shows that task-relevant components are sparse and partially disjoint, supporting structured component-level filtering as an effective way to reduce merging interference. Extensive experiments across vision (ViT, 20 tasks) and language (RoBERTa, 8 tasks; Qwen2.5, 4 tasks) benchmarks demonstrate that CASS improves a range of representative merging baselines.

Despite advances in test-time scaling and diffusion finetuning, guidance for Auto-Regressive Diffusion Models (ARDMs) remains underexplored. We introduce an amortized framework that augments a pretrained ARDM with an offline-trained controller. By previewing future rollouts, the controller learns stepwise corrections that anticipate observations under a terminal-cost objective, yielding a reusable policy for guided generation. Motivated by a stochastic optimal control view of ARDM trajectories, our method injects small controls within each denoising sub-step while staying close to the pretrained dynamics. We study this approach for data assimilation (DA) in chaotic spatiotemporal partial differential equations (PDEs), where existing methods are often computationally expensive and susceptible to forecast drift under sparse observations. At inference, DA becomes a feed-forward rollout with on-the-fly corrections, achieving an order-of-magnitude speedup over strong diffusion-based baselines. Across two canonical PDEs and a compact ECMWF Reanalysis v5 (ERA5) pilot spanning six observation regimes, our method consistently improves stability and accuracy over state-of-the-art alternatives, with similar improvements observed in a larger-scale GenCast study.

Inverse design of physical systems (molecules, devices, microstructures) often reduces to optimizing a high-dimensional structure against an expensive black-box simulator. Direct search is difficult because the space is non-Euclidean, feasibility is hard to encode, and each evaluation is expensive. We present Co-PiLOT, a latent optimization approach that maps candidates through a generative encoder--decoder, uses the decoder as a learned validity prior, and searches the latent space with physics-informed black-box optimization. The framework is applied on the inverse design of magnesium alloy microstructure/texture. We develop a vision transformer based--encoder; paired with latent diffusion, diffusion transformer and rectified-flow transformer--based decoders on $\sim80{,}000$ EBSD-derived microstructure dataset to learn a minimal bottleneck, $z$. The ViT-FMDiT model ($z$=$768$) reconstructs high-fidelity microstructure images (FID $23.19$, MS-SSIM $0.178$), which our self-segmenting orientation codec converts into input grids for crystal plasticity solver. Finally, we introduce Meridian, an active latent optimizer driven by deep-kernel Gaussian-process uncertainty, feasibility prediction, active trust regions, and target-aware acquisition. Within the evaluation budget, the ViT-FMDiT and Meridian combination yields the best target-driven objective score, outperforming DANTE, TuRBO, and BAxUS by $\sim6$\% in relative error on the same decoder.

Automatic neuron reconstruction from light microscopy images is a central problem in computational neuroanatomy. While recent methods have achieved encouraging results on local image blocks, it remains unclear whether such progress translates to reconstruction that is both structurally accurate and scalable to the whole-brain scale. We present CORAL, the first benchmark for structure-aware evaluation of automatic neuron reconstruction from light microscopy images at both local and whole-brain scales. Built on a high-quality whole-brain fMOST dataset with carefully curated annotations, CORAL establishes two progressive tasks: block-level reconstruction, which evaluates reconstruction methods under limited spatial context, and brain-wide reconstruction, which assesses complete neuron reconstruction at the whole-brain scale. To account for topological correctness beyond geometric distance similarity, we introduce a structure-aware metric based on fiber prediction. To further achieve complete neuron reconstruction across the entire brain, we develop a brain-wide neuron tracing framework that extends arbitrary local reconstruction methods to the whole-brain scale through an iterative local-to-global process. Using this benchmark, we provide the first structure-aware comparison of mainstream methods for local neuron reconstruction and further evaluate their performance in brain-wide reconstruction. Our results underscore the importance of structure-aware evaluation and the need for more robust methods for complete neuron reconstruction. All datasets and code are available at https://github.com/yang-ze-kang/CORAL.

A hallmark of neurodegenerative diseases such as Alzheimer's and Parkinson's is the aberrant aggregation of proteins into amyloid fibrils, and small molecules that selectively bind to these fibrils hold promise as diagnostics, imaging probes, and therapeutics. Predicting how such ligands bind to fibril targets, however, presents two fundamental challenges. First, resolved co-crystal structures of amyloid–ligand complexes are exceptionally scarce; even with recent advances in cryo-EM only a handful have been structurally characterized, making supervised training of docking models impractical for this target class. Second, amyloid fibrils present a binding mode fundamentally different from globular proteins: ligands intercalate into longitudinal cross-β grooves and stack cooperatively along the fibril axis, a geometry that existing docking models are not designed to capture. To address these challenges, we present CORAL (Cooperative Amyloid Ligand docking), a reinforcement learning framework that trains a generative docking model to produce ligand pose distributions tailored to the cross-β groove geometry. Our reward explicitly incorporates cooperative ligand–ligand stacking energy alongside protein–ligand docking affinity, directly capturing the distinctive binding geometry of amyloid fibrils. We further introduce a curated evaluation set of amyloid–ligand complexes constructed from model-generated poses validated by domain experts. Experiments on both experimentally resolved structures and this evaluation set demonstrate improved pose quality and binding affinity correlation over existing docking baselines.

Large language models have made knowledge-graph question answering more flexible, yet complex multi-hop questions often require evidence that becomes identifiable only during reasoning. Starting from the question alone, narrow retrieval can miss bridge facts, while broad expansion injects distractors and worsens long-context interference. The key bottleneck is dynamic evidence identification: the system must revise a compact evidence view as reasoning exposes unsupported facts. We formulate this process as partially observable control over evidence and propose CoRE, a residual-coupled framework in which reasoning trajectories and evidence subgraphs evolve together. CoRE decomposes each trace into answer-critical claims, treats unsupported claims as residual signals, and uses them to guide targeted retrieval and budgeted subgraph projection. The same formulation exposes a learnable projection decision: after residuals retrieve candidate facts, CoRE-RL learns from rollout feedback which facts should remain visible for the next reasoning pass. Across WebQuestionsSP and Complex WebQuestions, the CoRE family improves answer accuracy, evidence sufficiency, and budget robustness, with CoRE-RL providing further gains through more effective retained-subgraph selection.

Human labels are important supervisory signals for preference-based reinforcement learning, but their collection is often constrained by practical considerations that standard training objectives tend to ignore. For instance, Direct Preference Optimization (DPO) usually treats observed preference labels as clean, even though real annotation pipelines rely on heterogeneous annotators whose reliability varies across individuals and with the difficulty of each comparison. This setting introduces three challenges: (1) Noisy labels introduce systematic bias into the preference optimization objective and need to be rigorously corrected. (2) The allocation of pairwise comparisons to different annotators is a design problem that directly affects the statistical efficiency of the training pipeline. (3) Non-random routing creates selection bias in annotator evaluation, complicating downstream decisions such as compensation. We address these challenges within a unified framework for DPO that combines annotator-aware posterior correction with information-guided annotator routing. Posterior correction leverages an instance-dependent noise model to infer latent ground-truth preferences from noisy labels, mitigating bias in the DPO objective. Annotator routing is formulated as an optimal experimental design problem that allocates comparisons to annotators according to their expected statistical information under a fixed annotation budget. To support reliable annotator evaluation under non-random routing, we introduce a doubly robust estimator that corrects for selection bias. Semi-synthetic experiments on the UltraFeedback dataset show that our approach improves ground-truth preference recovery and downstream AlpacaEval 2 performance over baselines, with consistent gains across multiple open-source LLM model architectures. Annotator reliability-estimation experiments further show that the doubly robust estimator substantially reduces mean squared error relative to other baselines, demonstrating its value for annotator evaluation.

Correspondence pruning—separating sparse inliers from severe outliers in noisy feature matches—is a long-standing bottleneck for two-view geometry estimation in 3D vision. Existing learning-based pruners pursue this goal through ever more sophisticated context operators—local graphs, dynamic neighborhoods, or global attention—yet a controlled comparison reveals a striking phenomenon: when we freeze the structural assignment of these operators after a single forward pass, accuracy collapses by over $30\%$ mAP regardless of operator choice. This suggests that the limiting factor of current methods is not which operator is used, but the implicit assumption that structure can be committed in one shot. We argue that correspondence pruning should instead be cast as \emph{iterative structural rectification} (ISR), in which the geometric structure and the feature representation must \emph{co-evolve} layer by layer. Building on this view, we propose \textbf{ISR-Net}, a minimal architecture that pairs a layer-wise re-derived local-context operator (a dynamic $k$-NN graph in feature space) with a layer-wise re-derived global-context operator (soft assignment to learnable cluster prototypes), so that any gain over prior arts is attributable to the rectification \emph{schedule} rather than to operator novelty. Despite using deliberately off-the-shelf operators, ISR-Net attains state-of-the-art results on YFCC100M, SUN3D, and HPatches, surpassing the strongest attention-based competitor by $+8.2\%$ mAP@$5^\circ$ while running $35\%$ faster, indicating that \emph{how often structure is refined} matters more than the operator's complexity.

Crowd counting is a vital computer vision task with wide-ranging applications, from public surveillance to urban planning. A persistent challenge is achieving scale invariance despite noisy labels, as humans often make counting errors with similar objects. Most existing methods struggle to maintain consistency and robustness when testing images differ in scale from the training data. We propose ScaLe-aware Invariance and Correcting Errors (SLICE) for robust crowd counting, a novel scale-aware loss function that improves scale invariance, enabling consistent and robust predictions under varying scales and annotation noise. The loss function combines two components: a scale-invariance term that applies multiscale supervision to enforce consistency across density maps of the same scene, and an error correction term that addresses inevitable annotation noise by modeling label uncertainty with probabilistic methods such as a mixture of Gaussians. SLICE is model-agnostic and integrates easily into existing crowd counting architectures without extra computational overhead during inference. Extensive experiments on five benchmark datasets show that SLICE greatly improves scale invariance and cross-dataset generalization, achieving the state-of-the-art results in challenging scenarios. Using CrowdDiff, Steerer, and CrowdFormer as baselines, SLICE lowers mean absolute error (MAE) by 5–7\% by addressing annotation noise and scale variation. Applying the scale-invariance term across multiple scales further cuts MAE by up to 15\%.

Camera-conditioned video generation requires positional encoding that remains reliable under changes in camera motion, lens configuration, and scene structure. However, existing attention-level camera encodings either provide ray-only camera signals or rely on pinhole camera geometry, limiting their applicability to general camera control under the Unified Camera Model, including wide-angle and fisheye lenses. To address this limitation, we propose Curved Ray Expectation Positional Encoding (CRePE). CRePE represents each image token as a depth-aware positional distribution along its source ray, providing a Unified Camera Model-compatible positional encoding that naturally captures the curved-ray geometry of wide-angle and fisheye cameras. CRePE is implemented through a Geometric Attention Adapter added to frozen video DiTs, injecting token-wise scene-distance information into proper attention layers and stabilizing it with pseudo supervision from a monocular geometry foundation model. This design leads to more stable camera control and improved video quality in camera-conditioned video generation under the Unified Camera Model. Controlled positional-encoding ablations show consistent improvements over existing multi-view positional encoding, demonstrating the effectiveness of curved-ray-aware positional encoding across diverse camera models. Furthermore, by extending the same positional-encoding pathway to external geometry control through Radial MixForcing, CRePE supports external radial-map control for scene-geometry-conditioned generation and source-video motion transfer beyond camera control.

Simultaneous Speech Translation (SimulST) requires balancing translation quality and latency. While Large Language Models (LLMs) excel in offline translation, current LLM-based SimulST systems predominantly rely on prefix-incremental generation. This paradigm suffers from redundant Key-Value cache recomputation and necessitates external Voice Activity Detection to process unbounded streams. To overcome these bottlenecks, we propose an end-to-end interleaved read-write framework built upon an omni-modal LLM. By interleaving speech chunks and text tokens, our architecture achieves full KV cache reuse, natively supporting segmentation-free, infinite-length inference. Furthermore, to learn optimal streaming policies without rigid offline alignments, we introduce a novel Actor-Critic self-enhancement pipeline. An RNN-T Actor dynamically samples read-write trajectories, while an LLM-based Critic evaluates these paths via intrinsic semantic and latency rewards, distilling the optimal policy back to the Actor. Extensive experiments on sentence-level (FLEURS, CoVoST2) and document-level (MuST-C, ACL 60/60) benchmarks demonstrate that our model establishes new state-of-the-art quality-latency Pareto frontiers, substantially outperforming existing cascaded and static-alignment baselines.

Large language models from different families use different hidden dimensions, tokenizers, and training procedures, making behavioral directions difficult to compare or transfer across models. We introduce an anchor-projection framework that maps hidden representations from each model into a shared anchor coordinate space (ACS). Behavioral directions extracted from source models are projected into ACS and averaged into a canonical direction. For a new model, the canonical direction is reconstructed into its native hidden space using only anchor activations, without fine-tuning or target-specific direction extraction. We evaluate five instruction-tuned model families and ten behavioral axes. We find that same-axis directions align tightly across the Llama-Qwen-Mistral-Phi cluster on the ACS. This shared structure transfers to downstream tasks: held-out targets achieve \(0.83\) ten-way detection accuracy and \(0.95\) mean binary AUROC, while canonical steering induces refusal-rate shifts of up to \($\Delta$ = +0.46\) under distribution shift. Sensitivity analyses show that two source models and small anchor pools already suffice to approximate transferable directions. Overall, ACS provides a novel perspective on cross-family interpretability, revealing that representation-level transfer remains robust across model families. The code for this paper is available at \url{https://anonymous.4open.science/r/ACS-annoymous/}.

Modern machine learning progresses through empirical work, benchmarking new methods to evaluate relative performance. However, the statistical variability inherent to evaluation -- exacerbated by the stochastic nature of many algorithms -- often makes performance estimation unreliable due to the limited test samples available, leading to a validation crisis in which genuine advances are difficult to discern. In this work, we show that cross-validation improves markedly confidence when evaluating and comparing learning algorithm performances. We introduce the concept of \emph{sample gain}, which quantifies the virtual data augmentation achieved by using multiple cross-validation splits to reduce benchmarking variance. Experiments on both synthetic and real-world datasets (histopathologic scans and NLP fine-tuning) demonstrate that multiple splits can substantially improve the reliability and stability of performance estimates, with diminishing returns often setting in later than expected. We also introduce a procedure to dynamically early-stop cross-validation by estimating from the first few folds if subsequent folds will bring large sample gains. Our findings highlight the value of pushing cross-validation on available samples to achieve robust and reliable benchmarking.

Large Vision-Language Models (LVLMs) have achieved remarkable multimodal reasoning capabilities but remain critically vulnerable to adversarial perturbations. In practical black-box scenarios, severe architectural discrepancies, misaligned feature spaces, and divergent multimodal fusion strategies often undermine adversarial transferability, preventing perturbations crafted on surrogate models from generalizing to unseen target models. Existing ensemble-based LVLM attacks attempt to mitigate this by integrating multiple surrogates, yet they often fail to bridge the structural generalization gap across heterogeneous LVLM families. In this work, we propose a novel framework that enhances cross-model adversarial transferability via model diversity enrichment and stochastic parameter sampling. We show that transfer success is fundamentally governed by cross-model gradient variance. To mitigate this, we introduce (1) a diversity-induced continuous surrogate manifold that captures cross-model decision variability, and (2) stochastic parameter sampling during optimization to approximate the gradient distribution of unseen LVLM models. We provide a theoretical analysis bounding the expected adversarial transfer gain by gradient variance and prove that stochastic sampling reduces adversarial overfitting to surrogate decision boundaries. Extensive experiments across multiple state-of-the-art LVLMs demonstrate substantial improvements in black-box transferability under strict evaluation protocols.

AI systems operating near the limits of their capabilities can maximize utility while minimizing the risk of error by abstaining or deferring questions they are not confident about to a human or a stronger model. Recently, large language models (LLMs) have been post-trained to estimate confidence in their own answers. To achieve this, methods reinforce confidence scores based on the correctness signal to each question-answer pair \emph{individually}, with rewards that encourage assigning confidence 1 to correct answers and 0 to incorrect answers. However, estimating confidence scores is useful in that it enables selecting only the most confident answers, to maximize the share of solved questions while accommodating the user's risk tolerance. Assigning confidence 1 to correct answers and 0 to incorrect ones is theoretically optimal, but unattainable in practice and detrimental for selective prediction with a non-zero risk tolerance: confidences pushed to 0 and 1 carry no information about how answers compare to one another. We therefore introduce \ours (Cross-question Abstention Reward Estimation), which computes rewards over a set of question-answer pairs: At each of several risk tolerances, the confidences across the set induce an abstention threshold, and each confidence is rewarded for lying on the correct side of it. Since the threshold itself comes from the set, every verbalized confidence is reinforced relative to other questions' confidence, not against its own binary correctness label. On math, \ours increases the coverage at 5\% risk (C@5) across MATH-500, GSM8K, and Big-Math-Digits from 0.8\% to 42.4\%, compared to existing pointwise correctness-matching rewards. When trained on multi-hop Wikipedia QA from HotpotQA, \ours improves C@5 in out-of-distribution general-knowledge, reasoning, and math benchmarks (including CommonsenseQA, GPQA, SimpleQA, and TriviaQA) from 9.9\% to 23.6\%.

While LLM-based coding agents have advanced to repository-scale engineering, existing benchmarks predominantly focus on software maintenance, overlooking the foundational capability of system construction. We introduce CSBench, a benchmark evaluating project-level Spec-to-Code implementation across core computer science domains (e.g., OS, compilers, databases). CSBench features 100 expert-curated tasks from top-tier university assignments within containerized environments. Evaluating 12 state-of-the-art LLMs reveals a stark performance disparity: even top-tier models like GPT-5.2 see success rates plummet from over 90% in algorithmic tasks to below 35% in system-level domains. Our analysis identifies a fundamental deficiency in systemic reasoning and construction knowledge as the primary bottleneck, where agents either terminate prematurely due to overconfidence or fail to translate execution feedback into valid architectural fixes, highlighting CSBench as a necessary tool to bridge this gap and advance autonomous agents toward deep computer science mastery.

LLM-based theorem proving in Lean 4 has advanced rapidly, but provers are difficult to leverage newly contributed lemmas, because retrieval skews toward popular foundational lemmas, leaving the right premises buried among hundreds of thousands of candidates. Existing neural selectors rank candidates by semantic similarity, with three intrinsic limitations: frequency-skewed retrieval, isolated pairwise ranking, and degradation on out-of-distribution queries such as competition problems and informal language. Given that mathematics is structured by dependencies rather than similarities, we restore this signal to both the training data and the model. We curate two underused dependency resources. We conduct the first systematic mining of 53 Lean Blueprint projects, a previously untapped corpus of 3,805 manually curated informal+formal nodes that capture rare, research-level premises absent from Mathlib4. We further pair it with the extraction of comprehensive, typed, multi-level Mathlib4 dependencies. We then introduce DAPS, a dependency-aware premise selector with a structural neighborhood encoder, a group-level contrastive objective, and a Mathlib-then-Blueprint adaptation procedure. DAPS reaches Recall@32 of 89.31 on Mathlib4-Heldout (+11.47 over the strong selector LeanHammer). DAPS holds its lead across both out-of-distribution benchmarks, and improves three general-purpose LLMs on the informal IMOProofBench. Dependency graphs, benchmarks, and code are available 1 and will be released under a permissive license upon publication.

Subgraph mining aims to discover meaningful subgraphs from a large graph given a seed node, with applications in community detection, functional module identification, anti-money laundering, and knowledge discovery. Existing methods typically adopt static approaches that cannot dynamically and intelligently adapt to diverse graph structures. Moreover, training models for subgraph mining faces inherent challenges including exposure bias and the difficulty of learning when to stop expansion. To address these limitations, we propose DASM (Dynamic Autoregressive Subgraph Mining), a novel two-stage framework that models subgraph mining as a sequential decision process. In the first stage, we employ step-wise supervised learning with multi-label objectives and asymmetric F1 loss to teach the model how to select correct next nodes. In the second stage, we apply Group Relative Policy Optimization (GRPO) to align the model with end-to-end mining quality, enabling it to learn optimal stopping decisions through trajectory-level feedback. Our dynamic autoregressive approach allows the model to make state-aware decisions at each step, automatically determining subgraph boundaries without predefined size constraints. Experiments on multiple datasets demonstrate that DASM significantly outperforms existing baselines in terms of F1 score and IoU.

Fact-checking systems with search-enabled large language models (LLMs) have shown strong potential for verifying claims by dynamically retrieving external evidence. However, the robustness of such systems against adversarial attack remains insufficiently understood. In this work, we study adversarial claim attacks against search-enabled LLM-based fact-checking systems under a realistic input-only threat model. We propose DECEIVE-AFC, an agent-based adversarial attack framework that integrates novel claim-level attack strategies and adversarial claim validity evaluation principles. DECEIVE-AFC systematically explores adversarial attack trajectories that disrupt search behavior, evidence retrieval, and LLM-based reasoning without relying on access to evidence sources or model internals. Extensive evaluations on benchmark datasets and real-world systems demonstrate that our attacks substantially degrade verification performance, reducing accuracy from 78.7% to 53.7%, and significantly outperform existing claim-based attack baselines with strong cross-system transferability.

Decoding language intentions from intracranial electroencephalography signals enables direct communication for patients with severe speech impairments. However, practical deployment remains severely limited by cross-day performance degradation caused by physiological and instrumental heterogeneity, requiring patients to record extensive calibration data daily to maintain decoding performance. Here we introduce \textbf{He}terogeneity-Guided \textbf{D}ata \textbf{A}cquisition and \textbf{A}daptation (\textbf{HeDA\textsuperscript{2}}), a novel calibration paradigm that \textbf{determines which stimuli to record} for optimal calibration \textbf{before any neural data are acquired}, fundamentally departing from existing methods that passively adapt to whatever data happen to be available. HeDA\textsuperscript{2} operates through three stages: it first quantifies cross-day neural heterogeneity with a proposed \textbf{S}entence-level \textbf{H}eterogeneity via \textbf{A}lignment \textbf{P}ath \textbf{E}ntropy that captures how different phonetic phenomena exhibit varying degrees of heterogeneity; second, an estimator learns to predict this heterogeneity from historical patterns, enabling strategic selection of calibration sentences before any neural recording occurs; third, heterogeneity-aware adaptation updates the decoder using minimal acquired data. Extensive experiments on intracranial recordings across diverse recalibration settings demonstrate that, with merely 20\% of conventionally required calibration data, HeDA\textsuperscript{2} achieves comparable or superior decoding accuracy, representing a critical step toward clinically viable speech prosthesis.

Language model loss follows remarkably regular scaling laws over model and data size, yet it remains unclear why the aggregate loss should exhibit a power-law form. Existing explanations often attribute this regularity to a heavy-tailed spectrum of pattern difficulty in natural language, but this view has not been directly validated at token-level granularity in large-scale real-data training. We present a token-level framework that deconstructs scaling laws into localized learning events of individual contextualized tokens. By fitting token loss trajectories with sigmoids, we show that token learning is concentrated in localized transitions, giving rise to a learning-time spectrum that dominates the scaling-law shape. Across more than one hundred pre-training runs on large and diverse real-language corpora with modern LLM architectures, scaling up to 6B parameters and 300B training tokens, the measured learning-time spectrum quantitatively reconstructs the validation loss derivative along the training-step $T$, data-scale $D$, and model-scale $M$ axes. We further show that the same signal is actionable: by reshaping the training distribution according to when tokens become learnable, we alter the optimization trajectory and achieve 11\% faster validation-loss reduction. These results provide direct empirical evidence that scaling laws are governed primarily by the distribution of token-level learning times, and that this distribution can be used not only to explain scaling behavior but also to improve training performance.

Structural causal models (SCMs) provide a principled foundation for causal reasoning, and neural causal models (NCMs) extend this framework by parameterizing causal mechanisms with neural networks. The classical formulation of NCMs assumes a specific architectural choice: a single function approximator per structural equation. In this work, we revisit this assumption and show that it represents only one point in a broader family of valid neural causal model parameterizations. Using interventions as an analytical lens and graph-based neural architectures as a concrete construction, we characterize how different parameter-sharing schemes over sub-mechanisms induce distinct members of the NCM family while preserving the same causal graph. This perspective yields two additional regimes beyond the classical formulation, which differ in how causal mechanisms are decomposed and parameterized within a unified structural framework. Together, these regimes demonstrate that architectural choices shape the causal semantics and expressivity of neural causal models, even when graphical structure is fixed. Relaxing structural assumptions further, we identify a class of partially causal models (PCMs) that support limited interventional or counterfactual queries without satisfying full SCM requirements, offering alternative trade-offs between causal fidelity and tractability. We place NCMs and PCMs within a unified expressivity spectrum between purely associational models and fully causal generative models, and illustrate these distinctions empirically using a graph-based variational autoencoder with interventional structure.

Prompting steers large language models (LLMs) and vision--language models (VLMs) without weight updates, but it remains unclear how a change in instruction reshapes internal representations to produce a behavioral effect. We introduce a nested geometric decomposition framework that treats prompting as a transformation of the representational geometry for the content following the prompt. We ask what class of mathematical transformation best explains the effect of prompting by finding the best alignment between representations of the same stimulus set following different prompts. For each prompt pair, we fit a sequence of increasingly expressive stimulus-invariant maps: translation, rigid transformation with uniform-scaling, sequential axis scaling, affine, and nonlinear transformations. We then test these maps causally by replacing a single layer's prompt-A hidden state for a new set of stimuli with its mapped counterpart and measuring recovery of prompt-B representational geometry and behavior. Across three LLMs, three VLMs, and six text or image datasets varying in style, emotion, scene content, and number, prompts consistently reshape representational geometry toward the instructed task structure. In the cross-validated nested variance decomposition, much of the prompt-induced activation change is explained by shape-preserving maps: translation and rigid transformation with uniform-scaling. The tier profiles reveal model- and task-specific routing strategies, differing in how much transformation classes explain variance and where along the layer hierarchy their contributions emerge. Crucially, although translation and rigid transformation tiers already improve behavioral agreement, affine transformation is the first tier to nearly recover target-prompt task geometry and produces corresponding gains in behavioral agreement. This suggests that cross-dimensional linear mixing may be a key contributor to how prompts reorganize representations toward the instructed task structure. Our framework provides a general way to decompose prompt-induced representational change into interpretable geometric components, revealing how a model routes task-relevant structure to produce prompt-driven behavior.

Workload forecasting is a fundamental task to realizing electricity-computing synergy in modern data centers, where operational planning must account for the risk induced by fluctuating computing demand. Yet production workloads exhibit substantial volatility, making deterministic point forecasts insufficient for uncertainty-aware decision making. Existing methods mainly model workload uncertainty from temporal correlations in utilization traces, overlooking the job-induced cross-machine dependencies shaped by the scheduling layer. We empirically find that workload volatility exhibits structured and time-varying cross-machine patterns whose strength aligns with job co-occurrence. This observation motivates GD-Diff, a job-aware graph-conditioned diffusion framework for probabilistic workload forecasting. GD-Diff constructs a job-induced weighted dynamic graph from scheduling logs, transforming job co-occurrence into time-varying inter-machine dependencies. Built on this graph, GD-Diff uses a dual-branch noise prediction network within the diffusion process, where the temporal branch captures regular temporal evolution and the graph branch models job-induced dynamics. Experiments on Alibaba and Google cluster traces show that GD-Diff achieves state-of-the-art performance in both point and probabilistic forecasting, and ablation studies further validate the job-induced dynamic graph and dual-branch architecture. The code is available at \url{https://anonymous.4open.science/r/GD-Diff-F5E8/}.

Semi-supervised learning (SSL) relies on two core mechanisms: self-training under the Teacher-Student (T-S) framework and joint optimization of labeled and unlabeled losses. Despite their effectiveness, we find both mechanisms introduce distinct optimization pathologies. First, parameter coupling enforces strict synchronization between teacher and student, where strong regularization on the student degrades the teacher's fitting ability, thereby limiting the permissible generalization intensity. Second, the imbalance in gradient update consistency between labeled and unlabeled losses drives the shared parameters to prematurely converge to labeled-dominated local minima, creating a bottleneck for global optimization. To address both issues, we propose the Pioneer Student (PiS), an auxiliary branch that operates in an independent parameter space and periodically transfers accumulated knowledge back to the T-S model. Extensive experiments show that PiS is a universal plug-and-play module that consistently improves mainstream SSL methods.

Federated Spiking Neural Networks (FedSNNs) provide an energy-efficient and privacy-preserving learning paradigm for edge intelligence. However, their robustness under Non-IID data is limited by a coupled failure mode: label shift distorts membrane potential distributions, drives neurons away from gradient-sensitive regions, and further amplifies surrogate gradient errors, which destabilizes federated optimization. To address this problem, this paper proposes Spatio-Temporal Spiking Adaptive Thresholds (ST-SAT), a framework that mitigates the coupling between label shift and surrogate gradient errors through coordinated neuron-level adaptation and client--server knowledge alignment. At the neuron level, ST-SAT introduces the Spatio-Temporal Parametric Leaky Integrate-and-Fire (ST-PLIF) neuron, which adaptively calibrates firing thresholds, membrane potentials, and decay dynamics to keep local neuronal responses within effective gradient regions. At the collaborative interaction level, ST-SAT adopts a Dual-Scale Knowledge Distillation strategy that combines adaptive logit-level distillation with feature-level alignment of neuronal statistics, thereby reducing class distribution bias and improving knowledge consistency across heterogeneous clients. Extensive experiments on three benchmark datasets show that ST-SAT consistently improves both global model performance and client-local performance, with particularly strong gains under severe label shift. These results validate the effectiveness of the proposed framework for robust federated SNN learning. The anonymous code link is https://anonymous.4open.science/r/ST-SAT-4771.

Representation Autoencoders (RAEs) leverage frozen vision foundation models (VFMs) as tokenizer encoders, providing robust high-level representations that facilitate fast convergence and high-quality generation in latent diffusion models. However, freezing the VFM inherently constrains its spatial reconstruction capacity, limiting fine-grained generation and image editing; in contrast, incorporating reconstruction-oriented signals via fine-tuning disrupts the pretrained semantic space and degrades generative fidelity. To address this trade-off, we propose \textbf{DecQ}, a simple yet effective framework for RAEs. Specifically, DecQ introduces lightweight detail-condensing queries that extract fine-grained information from intermediate VFM features through condenser modules. These queries are incorporated into the decoder to support reconstruction and are jointly generated with patch tokens during generative modeling. By aggregating information from both shallow and deep layers, DecQ effectively mitigates the reconstruction--generation trade-off, improving both reconstruction quality and generative performance. Our experiments demonstrate that: (1) with only 8 additional queries and 3.9\% extra computation, DecQ improves reconstruction over the frozen DINOv2-based RAE, increasing PSNR from 19.13 dB to 22.76 dB; and (2) for generative modeling, DecQ achieves 3.3$\times$ faster convergence than RAE, attaining an FID of 1.41 without guidance and 1.05 with guidance.

The optimal transport (OT) map provides a geometric transformation for aligning probability distributions and has become a useful tool in machine learning. However, existing estimators of the OT map still exhibit a gap between sharp statistical guarantees and practical parametric estimation based on stable training objectives. Theoretical estimators achieve minimax optimal convergence rates, but they are typically nonparametric and can incur demanding implementation design or inference costs. Practical estimators are parametric and scalable, but their statistical guarantees remain underexplored, and their min-max, adversarial-like training objectives can be sensitive to optimization algorithms. We propose BROT (Barycentric Regression for OT), a simple two-step method that first computes the unregularized OT plan and then fits a deep neural network (DNN) to the induced barycentric targets by least-squares regression. Under standard regularity conditions, we prove that the DNN estimator of BROT attains the minimax convergence rate, when the ground-truth OT map is Lipschitz. Numerical studies on synthetic datasets and an image dataset show that BROT provides accurate map estimates, strong target distribution matching, and competitive transport costs, compared to existing estimation methods. Experiments on two downstream tasks, single-cell perturbation prediction and unsupervised domain adaptation, further suggest that the accurate estimation of BROT can translate into stronger task performance.

Turbulence is ubiquitous in engineering, yet direct simulation is prohibitively expensive. The Reynolds-averaged Navier-Stokes (RANS) equations provide computational savings exceeding ten orders of magnitude but introduce unclosed terms requiring modelling (the closure problem). Existing machine-learning (ML) closures suffer distribution shift when trained offline on high-fidelity data, and ML models that bypass the governing equations often have limited capability to generalise. We develop a physics-derived deep learning closure for RANS, the Deep Algebraic Reynolds Stress Model (DARSM), which can be trained on one or two flow cases and generalise across an order of magnitude in Reynolds number, to unseen geometries, and across flow regimes. A neural network maps flow invariants to empirical parameters in an implicit algebraic Reynolds stress equation, derived from the Reynolds stress transport equations under the weak-equilibrium assumption, imposing significant physics-based structure on the ML closure. End-to-end optimisation through the governing PDEs and the coupled implicit closure eliminates distribution shift, but both unrolled and implicit automatic differentiation fail on the stiff coupled solver. We derive adjoint equations that exploit the solver's implicit-explicit structure to enable computationally efficient optimisation. On the canonical square-duct and periodic-hill benchmarks, DARSM reduces test velocity error as compared to the baseline RANS by $2$-$4\times$ across Reynolds number, geometries, and flow regimes, with peak case-level reductions of $12\times$. The model trained on attached, anisotropy-dominated flows (square duct) accurately generalises without retraining to separated flows (periodic hills), a regime change in the underlying turbulence physics. DARSM also outperforms five established ML methods: offline training, tensor-basis neural networks, field-inversion machine learning, DeepONets, and physics-informed neural networks.

To account for uncertainty, hierarchical Bayesian optimization marginalizes over Gaussian process hyperparameters, usually averaging models or acquisition functions. However, this early aggregation discards information about model disagreement and is sensitive to outliers. To address these limitations, we propose to defer aggregation to the decision level: each model independently proposes a candidate, and we aggregate over these candidates via the medoid. We study a misspecified hierarchical Gaussian process setting in which most models have low misspecification error, while a minority are severely misspecified. Theoretically, we derive regret bounds showing that the misspecification penalty of early aggregation depends on the average error across all models, whereas decision-level aggregation depends only on the low-misspecification majority. Empirically, we validate these findings on toy examples and further demonstrate that, across a wide range of low- to high-dimensional synthetic and real-world benchmarks, decision-level aggregation outperforms standard Bayesian optimization baselines in most cases while remaining competitive in the others.

Building upon the real-time rendering capability of 3D Gaussian Splatting (3DGS), 2D Gaussian Splatting (2DGS) achieves improved surface reconstruction and geometric fidelity for novel view synthesis through explicit planar Gaussian primitives. However, the smooth falloff inherent to Gaussian kernels fundamentally limits the reconstruction of high-frequency shape signals, causing over-smoothed edges and loss of fine geometric detail. Prior works have explored alternative kernel functions to address this limitation, yet through systematic evaluation across signal fitting tasks, we find that all existing kernels suffer from the Gibbs phenomenon — producing oscillatory artifacts near sharp discontinuities. This observation motivates a fundamentally different strategy: rather than replacing the kernel, we make it deformable. This study presents Deformable 2D Gaussian Splatting, which augments each Gaussian surfel with a learnable, monotonic coordinate remapping in its local tangent plane. With only a single free control point per axis (two extra parameters), the remapping enables asymmetric radial profiles, allowing each primitive to reliably capture sharp edges and irregular structures that symmetric kernels inherently cannot represent. The deformation operates on the 2D tangent plane of 2DGS, preserving the differentiable rasterization pipeline with negligible overhead, and is optimized end-to-end through rendering loss alone. Experiments on MipNeRF 360 and Tanks&Temples demonstrate that our method effectively suppresses Gibbs artifacts at sharp boundaries and achieves state-of-the-art performance among 2DGS-based approaches in PSNR, SSIM, and LPIPS.

Multi-Task Learning (MTL) is a fundamental problem in machine learning that has been extensively studied over the past decade. Recently, a variety of optimization-based MTL approaches have been proposed to jointly learn multiple tasks by modifying the optimization trajectory. In this paper, we argue that the design of mainstream optimization-based MTL methods implicitly assumes compatibility with momentum-free optimizers, which may limit the understanding of their effectiveness in modern training regimes. In practice, the instantaneously derived gradients from MTL operations contribute only marginally to the final parameter updates when momentum is present, resulting in an amortized rather than instant de-conflicting effect. Moreover, this amortized behavior can be further degraded under high-curvature optimization dynamics. To address this issue and enable the effective integration of mainstream MTL methods with advanced optimizers (e.g., Adam and Muon), we propose \texttt{APT} (Applicability of advanced oPTimizers), a lightweight framework featuring an adaptive momentum mechanism that balances the trade-off between instant and amortized de-conflicting. Furthermore, we show that the Muon optimizer can be interpreted as an implicit MTL learner, and we introduce a lightweight direction preservation strategy to better align with its orthogonalization process. Extensive experiments across four standard MTL benchmarks demonstrate that \texttt{APT} consistently enhances existing MTL approaches, yielding substantial performance gains.

Reinforcement learning (RL) has become a prevalent paradigm for training tool calling agents, which typically requires online interactive environments. Existing approaches either rely on training data with ground truth annotations or require advanced proprietary language models (LMs) to synthesize environments that keep fixed once created. In this work, we propose TRUSTEE, a cost-friendly method for training tool calling agents with dynamic environments fully simulated by free open-source LMs that can be as small as 8B, including task generation, user simulation, tool simulation and trajectory evaluation, paired with an adaptive curriculum learning mechanism that controls task difficulty during training. Our empirical results show that TRUSTEE outperforms baselines which require extra external resources in most cases. These confirm that, with a sufficiently sophisticated design, even simulated environments with a local 8B LM as the backbone could set a strong baseline for tool learning. We hope our proposed paradigm could democratize tool learning and inspire future research on environment scaling with limited resources.

Time Series Anomaly Detection (TSAD) has received increasing attention, driven by the growing availability of complex time series data. This surge has led to the development of numerous detection methods, as well as a variety of benchmarks aimed at thoroughly evaluating their performance. However, most existing detectors remain largely agnostic to domain context, overlooking explainability and interpretability. One of the main reasons for this gap is that current benchmarks primarily focus on detection accuracy, and only few of them evaluate spatial explainability. Moreover, no benchmark currently provides sufficiently rich semantic annotations to support the generation of human-understandable interpretations of anomalies. To address these limitations, we introduce SHAD (Scality High-dimensional Anomaly Detection benchmark), a fully annotated benchmark composed of 144 multivariate, high-dimensional time series collected from real-world distributed cloud storage systems operated by Scality. The proposed dataset includes rich contextual information, covering three families of anomalies with varying degrees of severity. As further contribution, we provide an end-to-end benchmark that covers all stages of a TSAD pipeline. For Detection, we benchmark a wide range of existing anomaly detectors, testing their effectiveness on the proposed real-world dataset. Then, we evaluate explainability by measuring the contribution of each dimension in the generated anomaly score. Finally, for interpretability, we investigate the effectiveness of frozen LLM baselines in localizing and interpreting anomalies, paving the way for future research towards interpretable and explainable TSAD.

Accurately matching human intentions in 3D space is an important goal of artificial intelligence. Recently, 3D Intention Grounding (3D-IG) has emerged, aiming to localize target 3D objects given a natural-language intent. Unlike conventional visual grounding with descriptive referring expressions, 3D-IG intents are abstract and non-descriptive, making object localization substantially more challenging. This requires inferring latent functional requirements from non-descriptive intents and aligning them with object-level 3D representations. However, existing methods largely rely on implicit intent–object matching, leading to logical gaps and limited interpretability, robustness, and generalization. To address these challenges, we propose Chain-of-Causal Reasoning (CoCR), a causality-inspired functional dependency reasoning framework that explicitly bridges abstract intentions and candidate objects through intermediate functional requirements. Specifically, CoCR progressively decomposes complex intents into ordered functional requirements, thereby forming an explicit intent–function–object reasoning chain that links abstract intentions to object suitability. Building on this chain, we construct a causality-inspired functional dependency graph to model requirement--attribute relationships and introduce a causal-visual alignment module that aligns function-aware representations with the geometric-semantic features of 3D point clouds, enabling bidirectional verification between structured functional reasoning and visual evidence. Extensive experiments on 3D Intention Grounding and 3D Visual Grounding demonstrate that our method enhances intent-aware object localization.

Dexterous manipulation requires large-scale robot interaction data, yet collecting real-world demonstrations is costly, while sim-to-real transfer remains challenging due to visual and geometric discrepancies. We present DexMirror, a unified real-to-sim-to-real framework that enables photorealistic simulation learning and reliable deployment for dexterous manipulation. Our method reconstructs real scenarios using a compositional 3D Gaussian Splatting (3DGS) representation with tri-level optimization, obtaining high-fidelity objects and background while preserving global scene consistency. A dexterous-hand-centric calibration pipeline, supported by an interactive online platform, further achieves millimeter-level robot–scene consistency for efficient Gaussian-simulation alignment. Built upon the reconstructed scenes, we train privileged policies in simulation and distill them into visuomotor policies operating on rendered 3DGS observations, combined with closed-loop execution for robust transfer. Experiments demonstrate accurate real-to-sim reconstruction (F1 score 0.981) and strong sim-to-real performance, achieving high success rates across six real-world dexterous manipulation tasks.

Accurate forecasting of Earth's dynamic systems is vital in geophysics, where neural operators offer a powerful data-driven paradigm. Standard operators often treat the Earth as a flat Euclidean plane, introducing geometric artifacts. In contrast, Spherical Neural Operators (SNOs) resolve this mismatch by operating within the spherical spectral space, naturally adapting to spherical geometries. However, for computational efficiency, existing SNOs typically rely on the rotation equivariance assumption. This assumption simplifies the transitions of physical states into isolated spectral modal evolutions, thereby over-idealizing the Earth as a homogeneous system. To solve this problem, we no longer rely on the rotation equivariance assumption that isolates spectral modal evolution, but directly derive an explicit modal interaction kernel and propose Diagonal Spherical Neural Operators named DIAGNO. DIAGNO explicitly parameterizes cross-modal interactions through spectral modal decomposition, thereby preserving the inherent heterogeneity of Earth dynamics. Furthermore, leveraging the unique diagonal structure of the spherical spectrum, we design inter- and intra-diagonal interaction mechanisms to natively capture essential zonal and meridional dynamics, respecting intrinsic geophysical regularities. Extensive experiments across simulated fluid dynamics and real-world Earth system reanalysis (including atmosphere and ocean) datasets demonstrate DIAGNO’s superior capability, consistently achieving state-of-the-art performance in multi-step forecasting. Beyond this, we validate DIAGNO’s practical analytical value in professional geophysical domains by assessing its fidelity in capturing climatological anomalies and maintaining physical consistency.

We introduce DIAL (Directional Intervention, Asymptotically Limited), a parameter-efficient adapter that adds a bounded, monotone, scalar-controlled residual $\delta(s,c) = \sigma(g) \odot \tanh(c/T) \odot \tanh(Dz)$ with $z = D^\top \phi(s)$ to a base language model's hidden state. The architectural form yields a closed-form upper bound $\lVert \delta \rVert_\infty \le \sigma(g_{\max})$ on the residual $L_\infty$ norm, computable from model weights alone. Across 5 base models (3.8B to 70B parameters, 4 architecture families; one checkpoint per family, with 5-seed csweep variance at Llama-3.1-8B), the residual saturates at the predicted bound with mean ratio $1.000 \pm 0.001$ (worst-case family deviation 0.2 percentage points). On Llama-3.1-8B under GPT-4o judge ($n=100$ per bench, zero content-filter blocks), DIAL exceeds a FiLM baseline trained with the same multi-$c$ objective on every benchmark and judge. Semantic vs strict judge: HarmBench DIAL +0.52 / +0.55 vs FiLM +0.44 / +0.39; AdvBench DIAL +0.63 / +0.66 vs FiLM +0.46 / +0.50; XSTest DIAL +0.33 / +0.25 vs FiLM +0.26 / +0.22; both judges agree on the ranking. Outside training range, DIAL's residual saturates at $\sigma(g_{\max})$ while FiLM's grows unboundedly: at $c = 10^4$, FiLM's residual $L_\infty$ norm is 534 times larger than DIAL's. The bound holds across all control values, not only trained ones; concrete deployment scenarios (configuration error, multi-tenant policy ceilings, adversarial composition) are where this distinction matters in practice. The bounded-monotone form produces analogous saturation behavior on two continuous-control RL environments (HC-Vel $n=35$, Ant-Goal $n=30$) under matched oracle-$c$; the prior fails when $c$ must be inferred online (meta-RL). Ablations confirm both the bound and the saturating temperature are necessary; a frozen-gate variant retains the bound at lower controllability magnitude. We release adapter train reports and canonical bench generations, algebraic derivations (with a SymPy reproducibility script), the full GPT-4o judge cache (semantic and strict), and a human-labeled validation set ($n=50$, Cohen's kappa = 0.895). Full adapter checkpoints are deferred to camera-ready release.

Existing temporal graph clustering methods suffer from poor optimization dynamics due to reliance on heuristically initialized cluster assignment distribution without considering the dynamic nature of the evolving graph. The target cluster assignment distribution often conflicts with evolving temporal representations, leading to oscillatory gradients and unstable convergence. Motivated by the need for differentiable and adaptive clustering in dynamic settings, we propose TGRAIL (Temporal Graph Alignment and Index Learning), a novel end-to-end framework for temporal graph clustering based on Gumbel–Softmax sampling. TGRAIL enables discrete cluster assignments while maintaining the gradient flow. To ensure stable training, we formulate the clustering objective as an expectation over Monte Carlo samples and show that this estimator is both unbiased and variance-reduced. Furthermore, we incorporate a temporal consistency loss to preserve the order of interactions across time. Extensive experiments on six real-world temporal graph datasets demonstrate that our approach consistently outperforms state-of-the-art baselines, achieving higher clustering accuracy and robustness. Our results validate the effectiveness of jointly optimizing temporal dynamics and discrete cluster assignments in evolving graphs.

Text-to-image diffusion models have demonstrated remarkable capabilities in generating high-quality images, yet their tendency to reproduce undesirable concepts poses growing concerns for safe and controllable deployment, particularly regarding NSFW content, specific objects and styles. While existing concept erasure approaches primarily focus on DDPM-based diffusion models and rely on costly fine-tuning, the recent emergence of flow matching models introduces a fundamentally different generative paradigm for which prior methods are not directly applicable. In this paper, we propose Differential Vector Erasure (DVE), a training-free concept erasure method specifically designed for flow matching models. Our key insight is that semantic concepts are implicitly encoded in the directional structure of the velocity field governing the generative flow. Leveraging this observation, we construct a differential vector field that characterizes the directional discrepancy between a target concept and an anchor concept. During inference, DVE selectively removes concept-specific components by projecting the velocity field onto the differential direction, enabling precise concept suppression without affecting unrelated semantics. Extensive experiments on FLUX demonstrate that DVE consistently outperforms existing baselines on a wide range of concept erasure tasks, including NSFW suppression, artistic style removal, and object erasure, while preserving image quality and diversity.

At the core of reinforcement learning is the idea of learning beyond the performance in the data. However, scaling such systems has proven notoriously tricky. In contrast, techniques from generative modeling have proven remarkably scalable and are simple to train. In this work, we combine these strengths, by deriving a direct relation between policy improvement and guidance of diffusion models. The resulting framework, CFGRL, is trained with the simplicity of supervised learning, yet can further improve on the policies in the data. On offline RL tasks, we observe a reliable trend -- increased guidance weighting leads to increased performance. Of particular importance, CFGRL can operate without explicitly learning a value function, allowing us to generalize simple supervised methods (e.g., goal-conditioned behavioral cloning) to further prioritize optimality, gaining performance for "free" across the board.

We develop diffusion-based samplers for target distributions known up to a normalising constant. To this end, we rely on the well-known diffusion path that smoothly interpolates between a simple base distribution and the target, popularised by diffusion models. We tackle the score estimation problem by developing an efficient sequential Monte Carlo sampler that evolves auxiliary variables from conditional distributions along the path, providing principled score and density estimates for time-varying distributions. To control the variance of score estimates, we further propose practical control variate schedules that incur minimal overhead. We adapt this general framework to paths induced by the Ornstein–Uhlenbeck (OU) time-reversal process, stochastic interpolants, and diffusion annealed Langevin dynamics, outlining their trade-offs. Finally, we provide theoretical guarantees and empirically demonstrate the effectiveness of our method on several synthetic and real-world datasets.

We introduce a novel technique for scalable sampling of spin-system states with continuous symmetries using diffusion models. By applying our approach to the XY model, a fundamental continuous-spin model in condensed matter physics, we show that our technique addresses the shortfalls of the Markov chain Monte Carlo (MCMC) in generalization to varying system sizes. More specifically, we show that training a temperature-conditioned diffusion model on smaller-size XY model lattices enables the generation of accurate samples in larger lattice sizes. By tracking physically important observables of the model, such as spin correlations, our experiments demonstrate that diffusion sampling followed by few MCMC steps reduces the thermalization time by an order of magnitude relative to the standard MCMC with random initialization. Our study provides valuable insight as to how generative models can be used to study continuous-state condensed matter systems at scale.

3D shape completion from partial scans remains challenging for unseen categories and noisy real-world observations, where geometry alone is often insufficient for inferring missing structure. We present DinoComplete, a deterministic and efficient shape completion framework that augments geometric reconstruction with voxel-aligned semantic priors distilled from DINO features. First, we construct multi-view DINO feature volumes aligned with ShapeNet data and train a student network to predict dense semantic features directly from incomplete shapes. These predicted features capture global structure and part-aware semantic context while remaining aligned with the underlying geometry. We then integrate these distilled features into a completion network, where geometric and semantic voxel representations are fused through voxel state-space modeling. To enable efficient long-range reasoning without sacrificing resolution, we introduce a multi-scale voxel Mamba module that refines the fused features by combining full-grid and chunk-wise sequence modeling. Experiments on unseen ShapeNet categories and ScanNet objects show that DinoComplete achieves stronger completion quality than prior deterministic and generative based completion methods while using fewer parameters, requiring lower memory, and achieving faster inference. Our results demonstrate that distilling semantic priors from visual foundation models improves generalization and robustness in 3D shape completion.

We study nonparametric estimation of Schrödinger bridge (SB) drifts from i.i.d data observed on a single time interval. Starting from the conditional-ratio form of the Schrödinger bridge time-series (SBTS) drift formula, we analyze a direct Nadaraya--Watson plug-in estimator built from kernelized numerator and denominator terms. Unlike recent SB analyses based on entropic-OT potentials, Sinkhorn iterations, or iterative bridge solvers, our approach works directly at the drift level and isolates \emph{statistical error} from optimization, approximation, and discretization error. Under Hölder regularity, a marginal-density floor, and bounded support, we prove a uniform non-asymptotic bound for admissible bandwidth pairs, a pointwise CLT under genuine undersmoothing, and an adaptive bandwidth selector satisfying an oracle inequality. We also prove a pivot-local minimax lower bound which, through an explicit uniform pivot, yields a global minimax lower bound under transparent compatibility conditions; hence the adaptive selector is minimax-rate optimal up to logarithmic factors. Synthetic experiments provide theorem-targeted diagnostics for finite-sample scaling, Gaussian approximation, and adaptive behavior.

Humans and modern vision models can reach similar classification accuracy while making systematically different kinds of mistakes - differing not in how often they err, but in who gets mistaken for whom and in which direction. We show that these directional confusions reveal distinct inductive biases that are invisible to accuracy alone. Using matched human and deep vision model responses on a natural-image categorization task under 12 perturbation types, we quantify asymmetry in confusion matrices and link it to the shape of the information–error trade-off. We characterize this trade-off geometry using three signatures extracted from the rate–distortion (RD) frontier: slope (beta), curvature (kappa) and efficiency (AUC). We find that humans exhibit broad but weak asymmetries, whereas deep vision models show sparser, stronger directional collapses. Robustness training reduces global asymmetry but fails to recover the human-like breadth–strength profile of graded similarity. Mechanistic simulations further show that different asymmetry organizations shift the RD frontier in opposite directions, even when matched for performance. Together, these results position directional confusions and RD geometry as compact, interpretable signatures of inductive bias under distribution shift.

Conditioning diffusion models on visual features is essential for controlled image generation and counterfactual reasoning across diverse applications, from creative design to medical imaging. Existing conditioning methods, such as timestep embedding and cross-attention mechanisms, suffer from limited control over feature accuracy, require substantial architectural modifications, and struggle to maintain conditioning precision throughout the denoising process, as the model's focus shifts toward fine-grained details in later steps rather than the desired conditioned features. We present a novel approach that addresses these limitations by learning to shift the initial noise distribution in the latent space. Our method assigns each visual feature a distinct direction in the latent space and applies proportional shifts during both training and inference, enabling precise control over continuous and discrete visual attributes. We introduce a combined loss function that explicitly enforces shift accuracy alongside denoising, ensuring consistent conditioning across all denoising steps. Furthermore, we extend this framework to datasets with hidden or inaccessible visual features by employing a Variational Autoencoder to extract latent representations in an unsupervised manner. This enables high-fidelity counterfactual generation on complex, real-world datasets where explicit feature annotations are unavailable or prohibitively expensive to obtain.

Computational cognitive modeling has traditionally relied on manual model construction, which is a time-intensive process that requires substantial domain expertise. Although recent LLM-based approaches have begun to automate model generation, they often prioritize predictive accuracy at the expense of structural plausibility and parameter identifiability. We introduce DisCo, an LLM-guided framework that discovers interpretable symbolic cognitive models through evolutionary search with two structural components: a Validator that filters structurally invalid models, and a Cleaner that promotes parameter identifiability by removing redundant parameters. Across three behavioral tasks, DisCo consistently discovers models that achieve performance competitive with task-specific baselines. Ablation analyses show that the Validator is important for recovering psychologically meaningful structure. Without this component, models tend to violate behavioral constraints and show less consistent associations with external cognitive measures. These results suggest that LLM-driven model discovery, when coupled with appropriate structural constraints, can yield cognitively plausible models that go beyond predictive fit.

EHR foundation models create new opportunities for scalable clinical prediction across health systems, yet their deployment remains constrained by a basic tokenization bottleneck. The same clinical concept can appear as different ontology codes, local identifiers, synonyms, or free-text descriptions, while standard EHR tokenizers treat these variants as unrelated symbols. We introduce MEDIATOR, a discrete ontology-agnostic medical concept tokenizer that turns EHR harmonization into an open-form tokenization problem. Instead of relying on manual mappings, a fixed ontology vocabulary, or continuous event embeddings, MEDIATOR maps any clinical concept expressed as text into a shared vector-quantized codebook. MEDIATOR is trained with synonym supervision from paired concept descriptions, and further guided by enriched concept descriptions, empirical code co-occurrence, and curated relation graphs so that equivalent expressions collapse while non-equivalent concepts remain organized in a medically meaningful discrete space. The resulting vocabulary is compact, inspectable, reusable by standard EHR Transformers, and open to previously unseen code descriptions or text strings. Used as the token interface of an EHR Transformer, MEDIATOR improves mean AUPRC on 20 MIMIC-III tasks by 4.0% and improves zero-shot transfer to 14 eICU tasks by 8.8%, while supporting transferable attribution across hospitals.

Graph Foundation Models (GFMs) have attracted growing attention for their ability to model multi-domain graphs universally and generalize to unseen downstream tasks. However, existing GFMs typically rely on a single domain-invariant representation space, which inevitably blurs two fundamental connectivity patterns, \emph{homophily} and \emph{heterophily}, that govern how nodes relate across domains. Additionally, hard-partitioning the graph into disjoint subgraphs discards edges that are essential for self-supervised pre-training. In this work, we introduce $\underline{\text{HHGFM}}$, a $\underline{\text{H}}$omophilic and $\underline{\text{H}}$eterophilic Pattern-Disentangled $\underline{\text{G}}$raph $\underline{\text{F}}$oundation $\underline{\text{M}}$odel that encodes homophily and heterophily separately while keeping the original edge structure intact. Specifically, we propose a node-level soft disentanglement mechanism that estimates each neighbor's homophilic and heterophilic contributions and uses them as edge weights, thereby preserving the original graph structure. Low-pass and high-pass filters then serve as dual backbones to align homophilic and heterophilic patterns separately, supported by our generalization analysis. Extensive experiments on a range of graph datasets validate the effectiveness of our method.

The Platonic Representation Hypothesis (PRH) claims that independently trained models converge on a shared statistical model of reality, yet recent work finds only weak pointwise similarity between models. In this paper, we show that what models share is not the location of samples in representation space, but the directions (displacement vectors) between them. Under a single orthogonal alignment---rotation and reflection only---these displacement vectors are substantially preserved across 44 independently trained vision and language encoders spanning modalities and asymmetric capability pairs, consistent with the PRH evidence. The samples' absolute positions are not, consistent with recent counter-evidence. Both arise from a single decomposition: representations split into a shared semantic component that is linearly aligned across models, and a private capability component that is not. We trace this geometry to concept-level structure: within a model, parent concepts are orthogonal to their child variation vectors; across models, concept displacements are parallel. Our theory falsifiably predicts (and experiments confirm) that fine-tuning preserves pointwise similarity but collapses displacement, and that relational distillation does the opposite. A major implication is that, because semantics align linearly but capabilities do not, capabilities can be imported from one model to another using a single cached forward pass through the source. We call this Shadow Casting. As a proof of concept, our ShadowCLIP instantiation matches or outperforms strong fine-tuned baselines at orders of magnitude less compute. A cache can be released alongside open model weights, letting one model's capabilities be downloaded and imported into any number of other models without fine-tuning.

Modern generative models based on diffusion or flow matching transport noise toward the data manifold, without representing the data manifold explicitly. We propose \dm that explores an alternative design of generative modeling by explicitly modeling a distance field of the data manifold. Designing loss functions to learn distance fields in high-dimensional space from point data can be challenging. This is because the data do not provide reliable signals for closest point projection onto the manifold. To mitigate this problem, we design new loss functions that weight training data by their proximity to each sampling position. This distance-field perspective naturally motivates two principled deterministic samplers, showing fast convergence in inference. Our model is related to the recent time-unconditional generative models since it also does not require time input, and our analysis reveals that our method helps time-unconditional generative modeling to disambiguate the denoising problem and to avoid biasing towards the data mean. As a result, our model is the only time-unconditional model that remains competitive to time-conditional ones across datasets and architectures. Moreover, our distance prediction is also helpful for early stopping during sampling and for OOD detection. We hope distance-field modeling can open a promising direction for exploring generative modeling from a geometric perspective.

Conditional image generators learn $P(Y\mid X)$ for images given clinical, demographic, or experimental covariates, but their covariate effects are encoded implicitly in samples, denoisers, gradients, or score fields. We propose \emph{generative effect distillation} (GED), a framework that treats a conditional image generator as a teacher and projects its conditional law onto an interpretable image-response regression student $m_{\phi}(x,s)=\beta_0(s)+x^\top\beta(s)$ whose parameters are spatial effect maps. The teacher-to-map target is defined as a statistical functional of the teacher conditional law under a target design and discrepancy, rather than a set of synthetic images or a compressed generator. We develop mean, effect-gradient, and structured score distillation objectives; establish projection identities linking the distilled map to covariance-weighted teacher means and, under Stein conditions, to average teacher gradients; and derive finite-query risk bounds that separate teacher bias, spatial approximation, query design, Monte Carlo, and student optimization error. For the associated finite-basis coefficient-matrix oracle, a Gaussian lower bound matches the query-design and Monte Carlo stochastic dependencies. Across eleven blocks spanning synthetic, semi-real, pretrained SDXL-Turbo/FFHQ-256, CelebA conditional-diffusion, and face-image analyses, GED attains the lowest standardized integrated squared error (SISE) among the direct-regression and augmentation baselines in every block, with paired Student $t$-tests rejecting equality with the strongest non-GED baseline at $p<0.05$; on a CelebA conditional residual diffusion teacher, score-access GED via a single-step Tweedie denoising procedure attains SISE $0.009$, a factor of $25$ below the strongest non-GED baseline.

Time series anomalies take diverse forms, and individual detectors encode different anomaly scoring principles that rarely cover all anomaly types. Recent unsupervised time-series anomaly detectors provide complementary scoring signals, yet deploying them as an ensemble preserves the inference and maintenance cost of every individual detector. We study whether heterogeneous teacher detectors can be distilled into a single lightweight detector in the unsupervised setting, where no anomaly labels are available to define a shared target. AnoMix addresses this by extracting weak pairwise supervision from the training-set anomaly scores of frozen teachers. It normalizes each teacher score sequence independently and trains the student with CoRank, a multi-teacher ranking-distillation objective that converts teacher score gaps into pairwise supervision through teacher consensus. Across seven benchmark datasets, AnoMix successfully integrates complementary teacher scoring principles into a single deployable model, improving over individual detectors and static score ensembles while using a compact student with under 80k parameters on the benchmark settings. A zero-shot variant further shows that the distilled ranking signal transfers to unseen datasets and can outperform large pretrained time-series foundation models for anomaly detection.

Auction-based Federated Learning (AFL) provides a market-based mechanism to incentivize decentralized data owners (DOs) to join FL model training initiated by data consumers (DCs). However, optimizing the DC's bidding strategy remains fundamentally challenging. Existing methods predominantly rely on Reinforcement Learning (RL), which suffers from severe instability caused by the mismatch between stepwise Markovian rewards and the trajectory-level, non-decomposable feedback inherent in AFL. To address this mismatch, we transition from the stepwise RL formulation to a distributionally robust black-box optimization framework. By treating the entire recruitment-to-training process as a single high-fidelity function evaluation, this transition eliminates the need for explicit reward decomposition. We propose DR-AFL , which accounts for market non-stationarity and data distribution shifts by optimizing policies against worst-case environments within a Wasserstein ambiguity set. Our approach further incorporates an adversarial reweighting scheme to enhance robustness, alongside a dynamic safety projection that leverages real-time cost feedback to enforce budget-aware exploration. We provide theoretical analysis showing that DR-AFL induces a smooth surrogate over the discontinuous auction landscape and guarantees a robust performance lower bound under distributional shifts. Extensive experiments on 6 widely adopted datasets show that DR-AFL consistently outperforms state-of-the-art RL-based baselines, achieving a 2.43% improvement in model accuracy while reducing the number of training cycles required to reach target accuracy by 62.3%.

Learning a risk-sensitive individualized treatment rule (ITR) with patient covariates is challenging when the treatment criterion depends on the full conditional outcome distribution rather than only its conditional mean. Existing methods are largely criterion-specific, making it difficult to compare multiple treatment criteria within a common pipeline, such as quantiles, conditional value-at-risk (CVaR), and complex preference-based objectives such as cumulative prospect theory (CPT). We propose a distribution-first framework for risk-sensitive ITR learning that first estimates treatment-specific conditional outcome distributions and then constructs the target ITR by plug-in criterion evaluation. The proposed framework unifies treatment rule learning across multiple criteria and accommodates hard-to-optimize objectives within the same estimation pipeline. On the theoretical side, it yields a general error propagation result for Lipschitz treatment criteria, linking first-stage distribution estimation error to downstream criterion estimation and treatment regret, together with sharper regret bounds under standard margin conditions. Empirically, we show that the proposed framework outperforms direct criterion-specific methods in multiple settings and further evaluate it on the ACTG175 HIV clinical trial, a randomized study comparing several antiretroviral regimens in HIV-infected adults.

Distribution Matching Distillation (DMD) provides an effective distribution-level correction for few-step generation, while relying on an auxiliary fake-score network to track the evolving generative distribution. Recent work combines DMD-style objectives with flow-map generators to exploit both forward-divergence training and reverse-divergence correction. The fake-score estimator remains an additional component with memory and update overhead. In this work, we study whether this explicit tracker can be avoided when the generator itself has a flow-map structure. We propose \textit{Fake-Score-network-Free DMD (FSF-DMD)}, a DMD formulation for flow-map generators that replaces the auxiliary fake-score estimator with a generator-induced pseudo-velocity surrogate. The key observation is that the endpoint pseudo-velocity of a flow-map generator provides a tractable proxy for fake-velocity estimation, allowing the generator itself to supply the reverse-divergence signal. Building on this observation, we derive a practical objective, extend it with flow-map-consistent backward simulation, and introduce a self-teacher variant for training from scratch. In our ImageNet-1K $256\times256$ experiments, FSF-DMD improves flow-map baselines, reaches lower FID than the listed DMD2 comparisons in the flow-map-initialized setting, and remains effective under flow-matching initialization and training from scratch.

Generating high-fidelity mixed-type tabular data remains challenging because such data contain both numerical and categorical features; different columns often exhibit substantial heterogeneity in marginal distributions, sparsity patterns, and dependency structures. Although recent diffusion-based and flow-based methods have improved generation quality, most methods still rely on a unified generation framework that models all dimensions in a homogeneous manner, without explicitly accounting for the fact that different columns may differ in their suitability for strong local refinement. To address this issue, we propose a driver-navigator structured flow-matching framework (DN-Flow) for mixed-type tabular data generation. Inspired by the collaboration between the driver and navigator in rally car racing, DN-Flow explicitly decomposes the velocity field into a Driver branch (for globally stable transport) and a Navigator branch (for endpoint-aware local refinement). On top of this structured decomposition, we introduce a learnable column-wise soft selector and a dynamic correction control mechanism to jointly model column-level suitability for local refinement and the actual strength of correction injection in an end-to-end manner. Experiments on eight real-world datasets show that DN-Flow consistently outperforms strong baselines in data fidelity and downstream utility, achieves the best Shape performance on all eight datasets, and remains highly competitive on Trend and machine learning efficiency. These results suggest that structured and selective velocity correction offers an effective new paradigm for mixed-type tabular data generation. The code will be released upon acceptance.

Video understanding models achieve impressive 70%+ accuracy on recent benchmarks, but do these truly measure genuine grounded understanding? We systematically investigate this by controlling frame sampling from text-only to real-time across major benchmarks (VideoMME, Video-MMMU, LVBench). Our analysis reveals that 13–46% of questions exploit shortcuts through world knowledge, language priors, or single-frame cues, allowing models to answer correctly without engaging visual evidence. Furthermore, requiring temporal grounding, both correct answers and time localization, exposes a ~40% accuracy gap, showing models frequently answer correctly without locating relevant evidence in videos. To address these limitations, we introduce TGVMME, a temporally grounded benchmark with 1.5k questions requiring both answer correctness and temporal localization. Our benchmark reduces shortcuts to 3.5%, validating that temporal grounding effectively ensures genuine video understanding. Using TGVMME to analyze state-of-the-art reasoning models, we make a surprising discovery: inference-time reasoning provides minimal gains (+2.5% at best), contrasting sharply with substantial gains on existing benchmarks. Analysis on accuracy changes of shortcut questions confirms reasoning primarily exploits shortcuts rather than enhancing genuine understanding. Further experiments reveal input scaling (increasing frames) provides more benefit than output scaling (reasoning) given sparse inputs, highlighting the importance of frame-efficient architectures. Our work demonstrates that grounded evaluation is essential for assessing video understanding capabilities and provides practical tools: validated shortcut taxonomy, filtered splits, and TGVMME for rigorous evaluation.

Visual instruction tuning is crucial for improving vision-language large models (VLLMs). However, many samples can be solved via linguistic patterns or common-sense shortcuts, without genuine cross-modal reasoning, limiting the effectiveness of multimodal learning. Prior data selection methods often rely on costly proxy model training and focus on difficulty or diversity, failing to capture a sample’s true contribution to vision-language joint reasoning. In this paper, we propose CVS, a training-free data selection method based on the insight that, for high-quality multimodal samples, introducing the question should substantially alter the model’s assessment of answer validity given an image. CVS leverages a frozen VLLM as an evaluator and measures the discrepancy in answer validity with and without conditioning on the question, enabling the identification of samples that require vision-language joint reasoning while filtering semantic-conflict noise. Experiments on Vision-Flan and The Cauldron show that CVS achieves solid performance across datasets. On Vision-Flan, CVS outperforms full-data training by 3.5\% and 4.8\% using only 10\% and 15\% of the data, respectively, and remains robust on the highly heterogeneous Cauldron dataset. Moreover, CVS reduces computational cost by 17.3\% and 44.4\% compared to COINCIDE and XMAS. The code is anonymously available at https://anonymous.4open.science/r/CVS-ABAA.

Decentralized training of deep learning models is widely used to enable data privacy and on-device learning over networks. In realistic scenarios, heterogeneity across clients' local data poses an optimization challenge and can severely degrade test accuracy, especially on sparse communication graphs. Building on recent advances in centralized primal averaging, we propose Decentralized Primal Averaging (DPA), a decentralized learning algorithm that combines primal averaging with a local quasi-global momentum estimate. DPA maintains three coupled sequences for optimization, gradient evaluation, and output averaging, while using consecutive mixed iterates to construct a network-aware momentum direction without transmitting an auxiliary tracking variable. We prove a nonconvex stationarity guarantee for DPA's averaged output model and establish a linear-speedup result up to network- and heterogeneity-dependent residual terms. We also introduce DPA-1G, a one-gossip-per-iteration variant that matches the communication budget of DSGD-style baselines. Experiments following the GUT decentralized image-classification benchmark show that DPA outperforms DSGD, gradient tracking, QG-DSGDm, momentum tracking, and QG-GUTm across datasets, architectures, topologies, and heterogeneity levels, with the largest gains under high heterogeneity. DPA-1G matches or exceeds the one-gossip baselines at the same per-iteration communication budget.

Scalable Vector Graphics (SVG) are essential for modern industrial Computer-Aided Design (CAD). However, existing autoregressive SVG generation models are predominantly tailored for artistic creation and struggle to maintain the rigorous geometric fidelity and cross-view spatial alignment required for engineering drawings. To bridge this gap, we introduce $\textbf{DrawingsDreamer}$, a unified Large Language Model (LLM)-driven framework for multi-view vector-based engineering drawings generation. By formulating the generation of multi-view engineering drawings purely as a sequence modeling task, we eliminate the need of raster image encoders. We propose a Streamlined Representation utilizing hierarchical postfix tokenization, which guides the model to establish local geometric coordinates before assigning semantic boundaries. Optimized via a progressive task-aware curriculum schedule, $\textbf{DrawingsDreamer}$ effectively transitions from localized structural repair to macroscopic generation in a unified model. Extensive experiments demonstrate that our unified model achieves strong performance in both geometric fidelity and syntactic accuracy across diverse conditional and unconditional generation tasks.

Understanding user preferences from noisy and temporally evolving social media behaviors is fundamentally challenging due to interest drift, where user preferences shift across time and exhibit both multi-scale temporal patterns and diverse co-existing interests. To address this, we propose DUMoE, a unified framework for drift-aware multimodal user representation learning. Our model consists of (i) a temporal dynamics-aware backbone that captures and integrates static profiles, short-term behavioral signals, and long-term dependencies into a coherent representation, and (ii) a sparse mixture-of-experts (MoE) interest adapter that disentangles multiple latent interests via expert specialization and adaptive routing. Each expert models a distinct interest subspace, while a gating network dynamically selects and aggregates a sparse subset of relevant experts for each user. To enable stable and effective optimization, we further introduce a three-stage training strategy that decouples backbone learning, expert specialization, and gating optimization. Extensive experiments on real-world social media datasets show that DUMoE consistently outperforms state-of-the-art methods on both user interest prediction and interaction prediction tasks.

Driver attention prediction is crucial for interpretable driver monitoring and human-like autonomous driving. Existing methods typically formulate this task as a scene-to-heatmap regression problem, failing to reveal the underlying dynamic attention redistribution mechanism. To address this issue, we propose a scene-task-aware competitive allocation framework, which decomposes driver attention as a bottom-up visual saliency prior and a top-down scene-task utility representation. An adaptive product-of-experts allocation rule is designed to fuse these two heterogeneous branches in a competitive manner, where scene-task demand dynamically modulates the dependence on top-down task utility, and driving control urgency adjusts spatial attention concentration via a learnable attention budget temperature. To explicitly differentiate task-critical regions from visually prominent yet task-irrelevant distractors, we incorporate competition-aware contrastive learning to enhance feature discrimination in the latent space. Extensive experiments demonstrate that our method achieves strong and balanced performance, with a Kullback–Leibler divergence of 0.98 and a normalized scanpath saliency of 4.15 on the DR(eye)VE dataset, as well as a Pearson’s correlation coefficient of 0.67 and a histogram similarity of 0.54 on the BDD-A dataset. Ablation studies and diagnostic analyses further validate that our framework yields interpretable attention allocation behaviors, including urgency-induced spatial entropy contraction and scene-task-demand-guided dynamic attention reallocation.

Pretrained foundation models have become an important basis for end-to-end autonomous driving. In contrast to vision-language models pretrained primarily on static image-text pairs, video generative models capture temporal dynamics and motion priors that are naturally suited for driving. We present DriveWAM, a driving world-action model that adapts a pretrained video diffusion transformer into an autoregressive video-action policy. DriveWAM organizes video and action streams into a unified temporal token sequence and trains them under a joint flow-matching objective, preserving the pretrained video-generation architecture while adapting its large-scale video priors to action generation. To incorporate high-level scene understanding, we introduce scene-evolving driving guidance, where a frozen VLM produces chunk-specific semantic intent to guide video-action generation. To keep long-horizon rollout bounded, we further introduce selective KV memory, which maintains bounded modality-aware video and action memory pools through relevance-redundancy cache selection at inference time. Experiments on NAVSIM and the PhysicalAI-Autonomous-Vehicles benchmark show that DriveWAM achieves strong planning performance, and a data-scaling study from 4k to 100k driving clips further confirms the scaling potential of world-action modeling for end-to-end autonomous driving.

Preferential Bayesian Optimization (PBO) aims to find a decision-maker’s most preferred solution in as few pairwise comparisons as possible. Existing approaches rely on Gaussian Process (GP) surrogates, which provide strong performance but limited interpretability. This limits real-world usability in high-stakes domains, such as healthcare, where interpretability and trust are essential. We propose DT-PBO, a novel tree-based surrogate model for PBO that is inherently interpretable while capturing preference uncertainty. Specifically, we introduce a novel splitting heuristic that constructs interpretable shallow decision trees directly from pairwise comparison data, and use Laplace approximation to obtain probabilistic estimates within each leaf. This enables efficient preference modeling without sacrificing interpretability. Across eight benchmark functions, our method achieves competitive convergence to GP-based PBO, particularly on functions with rugged optimization landscapes. Additional experiments show robustness against noise and a fast computational running time. Experiments on four real-world datasets further demonstrate that our model provides interpretable insights into decision-maker preferences that remain opaque under GP-based approaches.

Drifting Models have recently achieved state-of-the-art performance in one-step image generation by training generators to follow corrective vector fields. However, their standard reverse-style drift improves samples from the model side, which can severely underutilize training samples and bottleneck early convergence. To address this, we introduce DualDrift, a unified framework that augments the original reverse drift with a novel forward drift. The forward drift lets each data sample assign correction to generated rollouts, providing denser data-side signals and accelerating early training. At the same time, this dense data-side signal can overemphasize dominant regions when used alone. DualDrift therefore uses a parameter-free scheduler driven by empirical batch statistics to balance the fast early correction of forward drift with the stable refinement of reverse drift. Evaluated on ImageNet $256\times256$, DualDrift achieves faster convergence and significantly improves over the baseline Drifting Models in both FID and Inception Score.

Transformers have become the dominant architecture for large language models, largely due to the scalability and flexibility of attention, feed-forward layers, residual connections, and normalization. This paper introduces dynamic short convolutions as an additional neural network primitive for improving Transformer architectures. Unlike static short convolutions, dynamic convolutions use input-dependent filters, preserving the locality bias of convolution while increasing expressivity. Motivating experiments show that applying dynamic short convolutions to key, query, and value representations improves performance on challenging associative recall tasks compared with static convolutional variants. Across language-modeling experiments ranging from 150M to 2B parameters, dynamic convolutions consistently outperform standard Transformers and Transformers augmented with static short convolutions. Scaling-law analysis indicates a 1.33$\times$ compute advantage over parameter-matched Transformers, while a custom Triton kernel enables efficient execution with minimal end-to-end slowdown. These results suggest that dynamic short convolutions are a scalable, hardware-efficient, and expressive primitive for advancing Transformer-based language models.

Model merging combines fine-tuned models into a unified one without joint training or access to original data. Dynamic merging improves flexibility by selectively activating task-relevant parameters, but existing methods either maintain a full shared model with tiny experts or allocate excessive capacity to experts, leading to suboptimal accuracy–efficiency trade-offs. We propose DiDi-Merging, a slim dynamic merging framework that uses differentiable rank allocation to balance shared and per-task expert parameters. The method casts parameter budgeting as differentiable rank optimization over low-rank modules, supervised data-free by the original task vectors, and applies a brief refinement stage to recover task fidelity. Across vision, language, and multimodal benchmarks, DiDi-Merging matches prior dynamic baselines at 1.24× the parameters of a single fine-tuned model and surpasses them at 1.4×, substantially more compact than methods requiring ≥2× storage.

Physical adversarial attacks expose important limitations of modern vision systems, but most existing methods rely on static patches or textures. We introduce a new problem setting: closed-loop physical adversarial control, where an agent interacts with a real-world perception system via a physical channel and receives only detector-level feedback. We optimize LED patterns with PR-SAC in a high-dimensional continuous action space. We show that latent action parameterization with deterministic upsampling improves training stability compared to direct control of individual LEDs. Across YOLOv8, Faster R-CNN, and RetinaNet, the learned patterns reduce person-detection confidence, decrease mAP@0.5, and exhibit partial cross-model transfer. We further analyze robustness under distance and viewpoint changes, interaction with a defense mechanism, and failure modes such as Q-drift, entropy collapse, and degradation after initial improvement. Our results suggest that dynamic physical attacks should be evaluated not only by best-case attack strength, but also by stability, transferability, and robustness under real-world variation.

Egocentric visual grounding requires high-resolution inputs to localize small objects. However, scaling Multimodal Large Language Models to this domain is constrained by the excessive cost of visual token processing. We identify that current efficient strategies based on token reduction are unreliable to select object-centric spatial evidence. To overcome this, we propose SmartRes, a framework that performs efficiency optimization in the pixel space via dynamic resolution routing. SmartRes first encodes a low-resolution view for global context and uses a lightweight router to activate high-resolution patches in object-centric regions and constructs an order‑preserving visual sequence. To further enable robust routing under severe foreground‑background imbalance, we introduce a margin-regularized routing objective that increases foreground-background logit separation and improves foreground recall. Experiments on Ego4D and EgoIntention show that SmartRes reduces visual tokens by up to 67\% while retaining 86.4\% of full-resolution performance, and achieves up to $1.66\times$ faster inference than state-of-the-art token reduction methods with higher accuracy. Furthermore, strong performance on small object grounding indicates the effectiveness of SmartRes towards egocentric applications. Code will be publicly available.

Medical vision-language models (VLMs) have shown increasing potential for clinical image interpretation, including lesion detection and report generation. However, their practical utility remains limited by insufficient sensitivity to subtle lesions, whose visual evidence is often sparse, low-contrast, and embedded within complex anatomical context. As local visual tokens are aggregated, these weak lesion cues can become underrepresented in global image representations, making them difficult for medical VLMs to recognize. Existing efforts to improve lesion sensitivity mainly rely on medical-domain vision-encoder pre-training, clinical-term-guided alignment, or trainable pathological representation enhancement. Although effective, these approaches usually require additional training or model-specific adaptation and may overfit to particular disease morphologies, limiting their applicability to frozen medical VLMs. To address these limitations, we propose \textbf{EasyLens}, a training-free plug-and-play subtle-lesion representation amplifier for medical VLMs. EasyLens first constructs \textbf{EasyBank}, a pathology-anatomy prototype space that provides lesion-related prototypes and anatomy-aware normal references for comparing suspicious patches against both pathological and normal anatomical patterns. To avoid blindly amplifying normal tissues, \textbf{EasyTag} selects lesion-relevant patches through counterfactual prototype reasoning. To counteract the dilution of subtle lesion cues in global image representations, \textbf{EasyAmplifier} strengthens the selected lesion-relevant patch representations through morphology-guided residual enhancement, thereby increasing their contribution to the global image embedding. Experiments on multiple medical image datasets and frozen medical VLM backbones show that EasyLens consistently improves subtle-lesion detection and outperforms existing encoder-enhancement baselines without model fine-tuning. Code is available at: https://anonymous.4open.science/r/easylens-BEC2

Deploying large language models on edge devices requires mitigating two major bottlenecks: the computational cost of the model and the memory overhead of the KV cache. While combining model quantization and KV-cache eviction seems like a natural solution, we reveal that this naive pipeline suffers from a critical failure mode. The issue stems from a hidden conflict: modern eviction methods rely on attention scores to identify important tokens, yet quantization inherently adds noise to these exact scores. Consequently, quantization does not merely degrade the numerical precision of the KV cache—it actively alters the eviction decisions, often causing the model to discard crucial context. To overcome this non-orthogonal degradation, we propose EAT, a training framework that integrates discrete eviction behavior into quantization-aware training. EAT computes eviction from quantized attention, applies the resulting mask in the forward pass, and uses a masked straight-through estimator in the backward pass. The method requires no architectural changes and adds no inference-time parameters. Systematic evaluations show that EAT mitigates the failure mode of naive quantization-plus-eviction pipelines. Under W4A16 + KV-INT8 with 50% KV eviction, eviction-unaware composition reduces the five-task average from 38.37 to 29.04, whereas EAT recovers it to 39.44 under the same cache budget. Real-world deployment on the MediaTek Dimensity 9500 demonstrates that, under a 50% KV-eviction budget, EAT improves decoding throughput by 64.1% and reduces total memory traffic by 20.4% compared to quantized full-KV inference. Mechanistic analyses suggest that the improvement is associated with reduced attention sink behavior rather than with sharper attention.

The availability of open 12-lead electrocardiogram (ECG) datasets has significantly increased the number of studies focusing on automated ECG analysis. However, existing datasets often have limitations due to unverified annotations, overly processed data, or the absence of features such as fiducial points of the ECG. In this paper, we present a comprehensive collection of ECGs, with each record meticulously annotated by a team of seven cardiologists. Over 2,500 ECG records were annotated, with most records reviewed by at least four physicians. 1,527 ECG records of the dataset contain detailed annotations of PQRST segments, verified by two cardiologists. The dataset constitutes the largest collection of manually delineated ECG records to date, offering a valuable resource for researchers and practitioners in cardiology. This dataset, the first of its kind available for open use with multi-expert annotations, is suited for a variety of research tasks. It can be used for ECG abnormality classification, for analyzing patterns and annotation complexities across different heart abnormalities, and for tasks related to ECG delineation.

Segmenting unseen anatomical structures in echocardiographic images is challenging because dense annotations are scarce and ultrasound often presents visually similar textures across different cardiac structures. Current foundation models remain limited in this setting because they treat anatomical knowledge as independent text prompts, overlooking the spatial topology that clinicians use to localize ambiguous structures from their visible neighbors. We propose Echo-SAM, a zero-shot ultrasound segmentation framework that grounds a structured cardiac knowledge graph in visual space through relation-conditioned geometric support inference. Echo-SAM localizes seen structures as anatomical anchors, propagates relational constraints through a knowledge-grounded Graph Neural Network, and enhances text prompts with image-grounded anatomical context. We further introduce a topology-to-geometry mapping that converts graph relations into geometry-aware support proposal scores, yielding posterior-guided support estimates for unseen structures. By treating relation-specific scale, thickness, and offset as latent geometric variables, Echo-SAM adapts graph-derived support proposals to image-specific anchors without relying on fixed coordinate templates. Extensive evaluations on zero-shot echocardiography segmentation benchmarks demonstrate that Echo-SAM substantially outperforms state-of-the-art foundation models, achieving up to 50.42\% absolute improvement in Dice score. Code will be publicly available.

Biological neural systems leverage neuronal heterogeneity and diverse firing patterns to generate rich nonlinear dynamics, yet these mechanisms are largely abstracted away in conventional artificial neural networks (ANNs) and simplified spiking neural networks (SNNs). Detailed differential-equation neural models are biologically expressive, but spike discontinuities and complex state evolution pose challenges for gradient-based optimization in complex functional tasks. To address this challenge, we introduce EDEN (Emergent Dynamics in Evolutionary Neural-Networks), a neural dynamic model training framework inspired by natural evolutionary processes. Built upon a continuous-time network of heterogeneous Izhikevich neurons, EDEN integrates efficient parallel neural dynamics simulation with advanced evolutionary strategies, enabling the joint optimization of intrinsic single-neuron-level parameters and network-level synaptic connections without backpropagation. Comprehensive evaluations across multiple MuJoCo continuous motor control tasks indicate that EDEN achieves robust control performance comparable to mainstream deep reinforcement learning baselines. Furthermore, neurodynamic analysis reveals that the trained EDEN models exhibit strong biological plausibility, characterized by functional differentiation across individual neurons and emergent low-dimensional latent manifolds governing population dynamics. EDEN demonstrates that the combination of gradient-free evolutionary strategies and biological neurodynamic networks provides an alternative yet efficient paradigm alongside mainstream reinforcement learning, thereby offering a highly promising pathway towards truly biologically-inspired intelligence.

Existing analyses of the edge of stability (EoS) treat it as a global property of optimization. We show that it is also selective: the stability constraint redistributes learning across subsets of the training distribution, amplifying progress on some groups while suppressing progress on others. Using a branching intervention that enters or exits the EoS regime from the same training state, we causally demonstrate this trade-off and identify two necessary conditions for a group to benefit. First, its aggregate gradient must align with the top Hessian eigenvector. We isolate this mechanism with a controlled perturbation that preserves distance but randomizes direction, destroying alignment and eliminating the advantage. Second, the group must sustain non-vanishing gradient magnitude over time. Under cross-entropy loss, gradient saturation decouples confidently classified groups, shifting the advantage to output-outliers, whose gradients persist. Together, these results show that EoS functions not only as a stability boundary, but as a mechanism governing the allocation of learning across the data distribution.

High-resolution image editing is increasingly demanded in professional workflows, yet existing diffusion-based models remain constrained to resolutions below 1K due to quadratic attention complexity and prohibitive memory requirements. A prevalent workaround employs a two-stage pipeline: editing at low resolution followed by independent super-resolution. However, this approach suffers from two critical issues: information divergence, where hallucinated details contradict the original high-resolution (HR) source, and texture degradation, manifesting as over-smoothed or over-sharpened artifacts. We propose EditBridge, a diffusion bridge framework for efficient ultra high-resolution editing. Unlike conventional diffusion that regenerates from noise, we formulate refinement as structured data-to-data translation from the low-resolution (LR) edited result to its HR counterpart, explicitly conditioned on the original HR source to preserve authentic details. To efficiently incorporate HR source guidance, we introduce a prior-guided block-wise sparse attention mechanism that exploits semantic correspondence from first-stage editing to constrain cross-image interactions to spatially aligned regions, significantly reducing computational overhead. Extensive experiments demonstrate that EditBridge achieves high-fidelity editing with superior perceptual quality at resolutions up to 4K, delivering 3.6--8.4$\times$ speedup at 2K and enabling practical 4K editing in 61 seconds.

Knowledge editing has become a key technique for updating knowledge in large language models. However, existing methods neglect the geometric structure of the knowledge representation space, leading to two key limitations: they cannot distinguish which pre-trained knowledge should be preferentially preserved (core vs. non-core) nor which new edits are inherently difficult (hard vs. easy). To address this, we propose \textsc{GEdit}, a \textit{geometry-aware} framework built on a Low-Rank Plus Diagonal (LR+D) factor model that captures the intrinsic low-rank structure of Transformer FFN hidden representations. \textsc{GEdit} features two complementary mechanisms. First, the \textit{low-rank regularization} applies targeted protection on a low-rank core knowledge subspace. Second, the \textit{adaptive regularization} dynamically adjusts the regularization strength based on each edit request's editability: relaxing constraints for hard edits to ensure success, and tightening them for easy edits to better preserve pretrained knowledge. Extensive experiments on \texttt{GPT2-XL}, \texttt{GPT-J}, and \texttt{LLaMA-3-8B} under large-scale editing scenarios—including both mini-batch and fully sequential editing settings with up to 10,000 continuous updates—show that \textsc{GEdit} consistently and significantly outperforms state-of-the-art baselines in both editing success and knowledge preservation.

Text-guided image editing with visual autoregressive (VAR) generators requires controlling both what the model samples and where the sampled change is written back into the image. Existing VAR editors mainly operate on token streams, features, or flat next-token logits, leaving two native structures of bitwise-residual VAR models underused: the per-bit Bernoulli prediction head and the additive multi-scale residual code field. We propose BitResEdit, a training-free editor for bitwise-residual VAR generators such as \textsc{Infinity}. BitEdit performs source-negative guidance by tilting the post-CFG per-bit log-odds along a source--target contrast computed on a shared edited prefix, then projects each update into a closed-form Bernoulli-KL trust region around the clean CFG sampler. ResEdit converts the resulting sampled bits into per-scale continuous-code residuals, gates them with a localization mask, and re-injects them through the generator's native sum-of-scales. The two components couple decision-time bit guidance with combination-time code composition, preserving masked-out latent features exactly while applying localized, scale-aware edits inside the target region. On PIE-Bench with Infinity-2B, BitResEdit achieves the strongest text alignment among same-backbone VAR editors and competitive background preservation, setting a new state of the art among training-free VAR editing methods. Ablations show that BitEdit and ResEdit play complementary roles in target alignment and background preservation. Detailed ablations show that both code-space residual editing and bitwise phrase-contrast guidance are necessary for robust autoregressive image editing.

Electroencephalography (EEG) foundation models increasingly rely on multi-dataset training and evaluation, yet public EEG datasets still lack a shared task specification layer that can turn heterogeneous recordings into reusable benchmark units. Existing standards organize files, metadata, and provenance, but they do not specify EEG tasks under a common language and rulebook, leaving critical task semantics scattered across papers, code, and manual interpretation. We investigate whether heterogeneous public EEG datasets can be standardized through a structured task specification language paired with a shared rulebook. Our methodology represents each benchmark entry as a task document synchronized with an executable task kernel, with the rulebook defining task fields, evidence requirements, document-kernel alignment, review states, and machine-checkable constraints. Using this methodology, we release a community-reviewed EEG benchmark corpus centered on 53 completed and reviewed entries with 245 task definitions spanning diverse paradigms, and we introduce NeuroDoc and NeuroAudit as the operational support layer for rulebook-guided drafting, upgrading, review, amendment, and release management. We further examine whether the resulting benchmark units can be instantiated in a shared downstream setting across four EEG foundation model backbones, providing execution-based evidence for reusable, auditable, and executable EEG benchmarking infrastructure. Project resources are available at https://anonymous.4open.science/r/anonymousrevieweneurodoc.

Knowledge conflict detection is a fundamental challenge for large language model (LLM) systems that rely on external knowledge. Existing approaches address this problem either with end-to-end supervised models or with multi-stage pipelines dependent on fixed heuristics. However, supervised models often generalize poorly to diverse conflict patterns, whereas heuristic pipelines rely on fixed rules and introduce considerable computational overhead. To overcome these limitations, we propose $\textbf{G}$rouping and $\textbf{C}$onflict $\textbf{D}$etection ($\mathsf{GCD}$), a two-stage agentic framework for knowledge conflict detection that aligns the detection process with the inherent sparsity of knowledge conflicts, thereby reducing the decision space and enabling fine-grained conflict detection. $\mathsf{GCD}$ decomposes knowledge conflict detection into two subtasks and employs two cooperative agents to address them: a Knowledge Grouping Agent (G-Agent), which creates conflict-preserving groups of textual units with potential conflicts, and an Intra-group Conflict Detection Agent (D-Agent), which detects conflicts within each group. The G-Agent's grouping policy is not fixed in advance but is jointly trained with D-Agent via a shared global reward, enabling grouping to be explicitly optimized for accurate detection rather than relying on fixed heuristics. Extensive experiments on five benchmarks demonstrate that $\mathsf{GCD}$ consistently outperforms SOTA methods, while reducing inference latency and token consumption.

Symmetric organization is a fundamental principle of biomolecular complex assembly. The same principle underlies engineered applications, from vaccine platforms to nanomaterials, making symmetric \textit{de novo} generation a core capability for computational protein design. We establish a theoretical foundation equating generation of an assembly with underlying symmetry group $G$ to generation within a quotient space over $G$. This directly allows $\mathrm{SE}(3)$-equivariant models to generate symmetric structures with minimal distribution shift. With this insight we introduce \textbf{\ours{}} (\textbf{Ze}ro-shot \textbf{U}nified \textbf{S}ymmetrization), a framework that converts any pretrained $\mathrm{SE}(3)$-equivariant backbone generative model into an efficient generator of symmetric assemblies with no retraining. \ours{} operates on a single \emph{canonical subunit} and utilize \emph{phantom subunits} to preserve internal model operations, reducing memory and runtime cost by $\mathcal{O}(|G|)$ for finite groups $G$, compared to a full-complex forward pass. We implement \ours{} for IPA-based models (FrameDiff, FoldFlow) and axial-attention-based RFdiffusion, showing high generation efficiency with minimal degradation to generation quality. We then apply \ours{} to generate symmetric complexes of over $10{,}000$ residues with exact, untruncated attention, a scale previously intractable with any model.

Efficient benchmarking techniques aim to lower the computational cost of evaluating LLMs by predicting full benchmark scores using only a subset of a benchmark's questions. By reframing this problem as an instance of multiple regression with feature selection, we find that existing efficient benchmarking methods can be greatly improved by simply using kernel ridge regression at the prediction stage. Additionally, using an information-theoretic feature-selection algorithm called minimum redundancy maximum relevance (mRMR), we can further improve upon these methods by selecting question subsets that will be maximally useful for prediction. Except in very data-poor settings, these approaches consistently achieve smaller prediction errors (in both MAE and RMSE), and greater ranking correlation between predicted and true scores (in both Spearman $\rho$ and Kendall $\tau$) across a range of benchmarks using both binary and continuous metrics. Furthermore, mRMR subsampling is much faster than competitor methods (which often involve fitting probabilistic models or running clustering algorithms), and is more likely to select the same questions under different random seeds or training data splits.

Text-to-Image diffusion models often propagate harmful bias inherited from the training data. Existing bias mitigation techniques either intervene only at the text encoder or provide inference-time guidance, often leading to generations that collapse into semantically incoherent outputs. To address these limitations, we introduce CO-ALIGN (Concept Ontology Alignment), a novel bias mitigation approach based on concept-graph alignment which operates on the model's internal concept ontology. By aligning concepts within the text encoder and denoiser, CO-ALIGN achieves significant bias reduction while preserving generative integrity. We demonstrate the effectiveness of concept graph alignment in three paradigms- text-encoders, denoisers, and joint text-denoiser ontology alignment. CO-ALIGN outperforms outperforms the state of the art, improving fairness by 30\%, $\Delta FID=11.4$ in image quality, 2.8\% in image fidelity, all while reducing semantically incoherent outputs by 88\%. Beyond bias mitigation, we show that CO-ALIGN benefits other downstream tasks as well. In particular, our experiments demonstrate that better-aligned internal ontologies enhance concept unlearning robustness across multiple unlearning techniques.

Exhaustively evaluating many large language models (LLMs) on a large suite of benchmarks is expensive. We cast benchmarking as finite-population inference and, under a fixed query budget, seek tight confidence intervals (CIs) for model accuracy with valid frequentist coverage. We propose *Factorized Active Querying* (FAQ), which (a) leverages historical information through a Bayesian factor model; (b) adaptively selects questions using a hybrid variance-reduction/active-learning sampling policy; and (c) maintains validity through *Pro-Active Inference*---a finite-population extension of active inference (Zrnic & Candès, 2024) that enables direct question selection while preserving coverage. With negligible overhead cost, FAQ delivers up to $5\times$ effective sample size gains over strong baselines on two benchmark suites, across varying historical-data missingness levels: this means that it matches the CI width of uniform sampling while using up to $5\times$ fewer queries. We release our source code and our curated datasets to support reproducible evaluation and future research.

LLM agents often execute long-horizon tasks where sequential tool calls and reasoning steps compound into a final outcome. When an agent errs mid-execution, the critical question is not whether an error occurred, but whether recovery under the current policy is still possible. Existing failure analysis answers this retrospectively, making it too late for runtime intervention. We introduce Task Failure Forecasting: predicting from a partial execution trace whether continued execution under the current agent policy is likely to fail. This is hard: human experts achieve only 54\% accuracy at distinguishing recoverable from failure-inducing errors, and frontier models such as Claude-4.5 Sonnet reach 59.87\% weighted F1. The core obstacle is that failed traces mix recoverable mistakes with fatal ones, obscuring which steps drive downstream failure. We address this with adaptive fault injection: injects targeted perturbations into successful traces, uses bandit-based prioritization of high failure yield error types and rollout-based verification providing policy-conditioned step-level supervision. A lightweight Qwen-3-8B model trained on this synthetic data achieves 78.01\% weighted F1 on human-annotated benchmarks, outperforming Claude-4.5 Sonnet by 14 weighted-F1 points at 100$\times$ lower forecasting cost. When deployed as a runtime monitor for selective inference-time intervention, the forecaster improves agent success rates across HotpotQA, MuSiQue, GAIA, AssistantBench, SWE-Bench, MBPP and EnterpriseBench, yielding absolute gains in the range of 2–16 \% points across held-out and unseen tasks.

Learning the natural parameters $z \in \mathbb{R}^n$ of discrete distributions $\mu_z$ from independent samples constrained to a subset $S \subseteq \\{0,1\\}^n$ is a foundational challenge in high-dimensional statistics. Existing methods for efficiently estimating truncated Boolean product distributions, notably the work of [Fotakis et al' COLT'20, Algorithmica '22], require either strong local connectivity assumptions on $S$ -- a property denoted *fatness* -- or stringent anti-concentration assumptions and necessitate the total mass of the truncation set to be a constant with respect to $n$. Moreover, the results in [Fotakis et al' COLT'20, Algorithmica '22] suffer from sample complexities that scale as $\Omega(2^n)$ if the mass of $S$ is exponentially small in $n$. In this work, we circumvent these limitations by analyzing the geometry of $S$ under the measure $\mu_z$. We refine the existing parameter estimation guarantees under the fatness assumption, improving the prior sample complexity to $\mathcal{O}( \log n / \epsilon^2)$ for $\ell_\infty$-recovery, matching the untruncated minimax rate. We further generalize fatness using the notion of influence utilized in the analysis of Boolean functions and provide sufficient conditions for efficient inference. Notably, unlike previous work, our method does not require sampling at arbitrary parameterizations of the model and instead relies on gradient descent. Lastly, we establish a theoretical lower bound demonstrating the sample complexity exhibits an intrinsic exponential dependence on the width of the model and the minimum distance between elements in the set.

Reinforcement learning (RL) post-training aligns diffusion-based generators with human preferences, yet existing RL methods suffer from poor compatibility with off-policy learning and few-step distilled models. These limitations are especially severe in the video generation area, as practical video generation pipelines often rely on few-step distilled generators. Furthermore, due to complex spatial-temporal dynamics and higher dimensions, near-on-policy video rollouts are both expensive to collect and often imperfect. Relying on such rollouts alone can amplify artifacts and is prone to reward hacking. To address these issues, we propose Forward-Consistent Reward Matching (FCRM), an efficient off-policy RL framework for video generation. FCRM converts the forward denoising loss into a positive loss-induced score and formulates the reward alignment as a one-step GFlowNet matching problem. The resulting residual is pointwise in a clean sample space that naturally supports off-policy learning and few-step generators. To avoid biased gradients, we introduce a double-sampling estimator for the squared residual objective. Theoretically, minimizing the proposed matching residual bounds the KL divergence between the learned distribution and the optimal reward-tilted distribution. Experiments on standard video generation benchmarks validate FCRM across online, replay, offline, and few-step settings and outperform SOTA methods.

Real-world image super-resolution aims to recover high-quality images from complex and unknown real-world degradations. However, existing generative Real-ISR methods largely inherit the dense latent representations and quadratic-cost global modeling paradigm developed for high-resolution image synthesis, causing computation, memory usage, and inference latency to scale unfavorably with resolution and thus limiting practical deployment. We argue that the key bottleneck lies not in insufficient restoration priors, but in excessive token redundancy and costly token interactions during high-resolution restoration. Motivated by this observation, we revisit Real-ISR from the perspectives of compact latent representation and linear-complexity modeling, and propose SANA-SR, an efficient one-step restoration framework. Specifically, SANA-SR employs a deep compression autoencoder with a 32× compression ratio to drastically reduce latent tokens while preserving restoration-relevant structures and textures. On top of this compact latent space, we introduce a linear-attention DiT with LoRA fine-tuning, enabling efficient high-resolution restoration with linear-complexity token mixing. Extensive experiments on all benchmark datasets demonstrate that SANA-SR achieves highly competitive and often superior quantitative performance against existing methods, while restoring clearer and more realistic textures. Moreover, after pruning, the deployed model runs in 0.019 s with 407.95G MACs and 344M parameters, highlighting its strong potential for practical mobile deployment.

LLM-based workflows compose specialized agents to execute complex tasks, and these agents usually share substantial context, allowing KV-Cache reuse to save computation. Existing approaches either manage KV-Cache at agent level and fail to exploit the reuse opportunities within workflows, or manage cache at the workflow level but assume that each workflow calls a static sequence of agents. However, practical workflows are typically dynamic, where the sequence of invoked agents and thus induced cache reuse opportunities depend on the context of each task. To serve such dynamic workflows efficiently, we build a system dubbed PBKV (\textbf{P}rediction-\textbf{B}ased \textbf{KV}-Cache Management). For each workflow, PBKV predicts the agent invocations in several future steps by fusing the guidance from historical workflows and context of the target workflow. Based on the predictions, PBKV estimates the reuse potential of cache entries and keeps the high-potential entries in GPU memory. To be robust to prediction errors, PBKV utilizes the predictions conservatively during both cache eviction and prefetching. Experiments on three workflow benchmarks show that PBKV achieves up to $1.85\times$ speedup over LRU on dynamic workflows, and up to $1.26\times$ speedup over the SOTA baseline KVFlow on the static workflow.

Touch is fundamental to dexterous manipulation, yet most egocentric human data increasingly used for robot learning lacks tactile information. Directly collecting large-scale tactile data is challenging due to sensor limitations, while human video data is abundant, contact-rich, and easily scalable. This motivates a natural question: can tactile signals be inferred purely from vision? To address this, we introduce EgoTac, a generalizable model that predicts rich tactile information directly from egocentric human videos. EgoTac is trained on a unified corpus of over 5.7M image-tactile pairs, covering both continuous force measurements and binary contacts. By learning from this diverse dataset, EgoTac captures nuanced touch dynamics across varied interactions. Experiments demonstrate strong performance: in-domain prediction achieves an average force error below 0.06N. On out-of-domain contact prediction benchmarks, EgoTac consistently outperforms the state-of-the-art contact estimator. It also captures the rise and fall patterns of real tactile data and enables zero-shot predictions on unconstrained real-world videos. Scaling analyses further reveal that both data diversity and volume improve performance steadily. Overall, EgoTac provides a scalable pathway to extract tactile priors from egocentric human videos, enabling broadly applicable tactile-aware robot learning.

Recovering global 3D human motion from monocular video captured by a moving camera is a fundamental yet challenging problem, as camera ego-motion and human body motion are tightly entangled in the image observations. The prevailing two-stage paradigm treats camera estimation and motion reconstruction as isolated processes, causing errors on both sides to be further amplified when combined in the world coordinate system. To address this, we draw inspiration from the human inner visual simulation mechanism and propose EIHMR, a collaborative human-camera co-estimation framework. EIHMR comprises two complementary modules that bridge scene-aware human motion refinement and motion-aware camera estimation: Scene-Aware Local Human Motion Reconstruction reprojects motion sequences into frozen keyframe viewpoints and leverages metric depth and kinematic constraints to produce geometrically consistent local motion, while Motion-aware SLAM re-renders the refined motion as static meshes in the original frames, converting dynamic human regions into structured matching cues for robust camera estimation. EIHMR consistently improves global trajectory reconstruction over strong baselines, demonstrating the effectiveness of collaborative human-camera estimation for long-range human motion recovery.

Large Language Models (LLMs) have demonstrated remarkable capabilities in Information Extraction (IE) tasks; however, their performance in zero-shot Named Entity Recognition (NER) remains suboptimal. We argue that this limitation stems not from intrinsic defects within the models, but rather from the failure of existing instruction paradigms to effectively elicit their latent capabilities. Drawing inspiration from cognitive science, we propose an instruction optimization framework termed Instruction Semantic Elaboration (ISE), which fully elicits models' latent NER capabilities by mimicking the semantic elaboration process—specifically by contrasting related concepts, decomposing constituent elements, and illustrating with concrete examples. We evaluate our framework under a zero-shot setting across four domain-specific datasets. Results demonstrate that ISE effectively and stably unlocks the intrinsic NER capabilities of LLMs, achieving a 10.99\% improvement in F1 score over initial instructions and outperforming strong baselines by 6.89\%. Furthermore, the framework generalizes well across model architectures and parameter scales, while remaining robust to the quality of initial instructions. This study offers a novel perspective on activating NER capabilities in LLMs, facilitating their efficient and stable deployment in real-world scenarios.

Modern AI is opening the door to collective decision-making in which participants express their views as free-form text rather than voting on a fixed set of candidates. A natural idea is to embed these opinions in a vector space so that the substantial literature on facility location problems and fair clustering can be brought to bear. But standard text embeddings measure semantic similarity, whereas distances in facility location problems and fair clustering require what we call preferential similarity: a participant's agreement with a piece of text should be inversely related to their distance from it. Off-the-shelf embeddings inherit a coarse preference signal through a correlation between semantic and preferential similarity, but fail to capture preferences when the correlation breaks. We formalize this as an invariance problem: text embedding models encode both a preference-relevant signal (stance and values) and semantic nuisance (style and wording), and the two are observationally correlated, so a geometry that relies on nuisance can appear preference-correct even when it is not. We show that synthetic training data designed to break this correlation provably shifts the optimal scorer away from nuisance-dominated cosine and significantly improves preference prediction across 11 online deliberation datasets.

Intrinsic Motivation (IM) aims to train agents without external rewards, enabling useful behavior to emerge from the agent's interaction with its environment alone. However, the dominant IM approaches rely on information-theoretic quantities with designer-chosen variables, introducing bias and lacking a principled connection to dynamics or optimal control (OC). We introduce Controllable Information Production (CIP), a new foundation for IM explicitly grounded in dynamical systems and OC. CIP measures the rate at which an agent produces information, capturing controllable complexity without external knowledge or bias. CIP unifies IM and OC into a single framework, formalizing physical intelligence as the control of information production. It further reveals connections between the structure of the value function and Kolmogorov–Sinai entropy. CIP consistently outperforms prior IM methods on standard benchmarks in robot learning and solves tasks they fail on, including humanoid self-righting. These results support a general organizing principle: physical intelligence emerges from driving systems toward the edge of controllable chaos.

Humans have a remarkable ability to judge causal relationships from a limited number of unreliable observations. Past work on causal cognition has largely focused on normative accounts of human behavior, leaving unknown how biologically plausible neural systems could learn causal relationships from observations and update their representations of causal structure with additional evidence. Here, we leverage task-optimized recurrent neural networks to discover candidate implementation-level neural mechanisms of causal judgment. We propose a novel cognitive task in which a subject observes stochastic samples from an unknown causal structure (e.g. among variables A, B, and C with unknown causal relationships), and must judge whether a specific causal relationship is present given a query (e.g. "does A cause C?"). We found that, after training, recurrent neural networks perform the task with high accuracy, adopt strategies that incorporate the behavior of non-queried variables to form their judgments, and, despite being trained only on pairwise queries ("does A cause B?", "does C cause A?", etc), form implicit beliefs about the complete graphical structure underlying the observations. Lastly, we use dynamical systems analysis to identify a set of low-level neural mechanisms that implement causal judgment and representation of causal graphical structure. Together, these findings lay the groundwork for a "bottom-up" approach to causal cognition, providing a potential basis for subsequent experimental study in the brain.

Facial expressions are one of the main ways humans communicate affect, but they are not static signals. In natural vision, expressions evolve over time, with facial movements emerging, intensifying, and changing as emotion is expressed. Prior model-brain studies based on static images suggest that vision models can capture aspects of face-related neural representations, but it remains unclear whether artificial vision models capture the temporal EEG dynamics of facial emotion perception. We test the hypothesis that models trained for facial emotion recognition, especially video-based or temporally structured models, should show stronger alignment with human EEG dynamics than randomly initialized controls. We do so by aligning model representations with time-resolved EEG representational geometry from 25 participants viewing one-second dynamic facial-expression videos. We compare six models spanning static-image, video-based, self-supervised, and vision-language representations: ResNet18, VGGFace, CNN2D, CNN3D, DINOv2-temporal, and Qwen2.5-VL. Contrary to this hypothesis, supervised facial-emotion training does not consistently increase alignment with human EEG dynamics. In particular, video-trained and temporally structured emotion models align less with fear-related EEG geometry than the same architectures with randomly initialized weights, whereas static-image emotion models show trained-versus-random differences close to zero. These results are robust to stricter noise-ceiling thresholds and are not explained by actor identity or raw pixel similarity. Exploratory analyses further suggest that peak alignment is often localized to the first or last sampled video frames, and that fear alignment varies across scalp ROIs by training regime. Overall, our results challenge the assumption that video-based emotion supervision automatically yields more brain-like dynamic representations of facial emotion.

While existing time series foundation models primarily rely on large-scale unimodal pretraining, they lack complementary modalities to enhance time series understanding. Building multimodal foundation models is a natural next step, but it faces key challenges: 1) lack of a unified multimodal pretraining paradigm and large-scale multimodal corpora for time series analysis; 2) how to effectively integrate heterogeneous modalities and enhance model generalization. To address these challenges, we take an early step toward multimodal foundation models for time series analysis. We first propose a multimodal pretraining paradigm that leverages time series with endogenous modalities (derived images and text) and exogenous knowledge (news), providing a comprehensive multi-view perspective for time series analysis. To support this, we develop an automated data construction pipeline to curate MM-TS, the first large-scale multimodal time series dataset spanning six domains, with up to one billion points. Then we propose HORAI, a frequency-enhanced multimodal foundation model. It integrates two core components: the Frequency-enhanced Cross-Modality Encoder and the Time-Frequency Decoder, designed to effectively fuse multimodal features and enhance model generalization across modalities and domains. After pretraining on MM-TS, HORAI achieves state-of-the-art zero-shot performance on time series forecasting and anomaly detection tasks, demonstrating strong generalization.

Modern 3D structure predictors rely on multiple sequence alignments (MSA) to expose evolutionary covariation, but for RNA those alignments are often missing, shallow, or unreliable. To overcome this limitation, we invert the co-evolutionary structure inference pipeline and encode a predicted 2D topology into synthetic alignment-form covariation, without requiring database search nor alignment steps. We introduce RNAformer, a transformer-based RNA secondary-structure predictor that supplies the topology, and a Synthetic Co-evolution Engine (SCE) that encodes it into synthetic homologous sequences (SHS) by co-mutating paired positions. Comprehensive experiments and systematic interventions across multiple structure predictors show that the MSA input acts as a topology channel that allows predictable steering of 3D structure prediction. Across single-chain RNAs and single-chain RNA-protein complexes, RNAformer-seeded SHS approach the performance of natural MSAs, with regime-dependent gains. Topology, not alignment depth, drives the response. SHS are interpretable, fast, and scalable, and represent a first step toward an alignment-free alternative to natural MSA.

Learning identifiable representations in deep generative models remains a fundamental challenge, particularly for sequential data with regime-switching dynamics. Existing approaches establish identifiability under restrictive assumptions, such as stationarity or limited emission models, and typically rely on variational autoencoder (VAE) estimators, which introduce approximation gaps that limit the recovery of the latent structure. In this work, we address both the theoretical and practical limitations of this setting. First, we establish identifiability of a broad class of recurrent nonlinear switching dynamical systems under flexible assumptions, significantly extending prior results. Second, we introduce $\Omega$SDS, a flow-based estimator that enables exact likelihood optimization using expectation-maximisation. Through empirical validation on both synthetic and real-world data, our results demonstrate that $\Omega$SDS achieves improved disentanglement compared to VAE-based estimators and more accurate forecasting of underlying dynamics.

We propose Energy-Based Operators (EBOs), an architecture-agnostic framework for learning conditional distributions over functions on continuous domains. An EBO defines a scalar energy over target functions given an input function and induces a probability model through a Gaussian reference. The resulting score field is obtained as the gradient of the parametrized energy to perform function-space iterative energy minimization (EM). Our model shows strong performance in function generation, such as super-resolution and forecasting, over various 1D function classes (oscillations, damping, and Izhikevich) in comparison with prediction operators and denoising operators over various architecture backbones. Moreover, it achieves strong performance over PDEs, namely Navier-Stokes, Darcy flow and Burgers. Notably, our model successfully detects anomalous functions by automatically assigning high energy without any supervision. It enables seizure detection and volatility prediction after learning neural dynamics and market microstructure dynamics without pre-defined labels during training, highlighting its effectiveness for both learning dynamical systems and detecting functional anomalies arising in scientific simulations.

Zeroth-order (ZO) optimization enables memory-efficient large language model (LLM) fine-tuning, with recent methods employing curvature-aware strategies. However, in the massive LLM parameter space, these methods inherently suffer from exploding gradient variance and prohibitive costs. Conversely, confining updates to a low-dimensional subspace reduces variance but discards critical high-dimensional information, degrading performance. This creates a fundamental dilemma: full-space optimization is informative but high-variance, while subspace optimization is low-variance but blind. To break this trade-off, we propose \textbf{\Engine} (\textbf{En}do\textbf{g}enous Var\textbf{i}atio\textbf{n}al MultiScal\textbf{e}). Inspired by computational mechanics, we adapt static Variational MultiScale theory to dynamic optimization, treating low- and high-dimensional spaces as resolved coarse and unresolved fine scales. Through a multigrid-style V-cycle, \Engine\ alternates between spaces, dynamically probing fine-grained information in the high-dimensional space and injecting it to correct the efficient low-dimensional sampling process. Evaluations on standard LLM benchmarks show that \Engine\ achieves higher accuracy and faster convergence than existing ZO methods, achieving up to a $\mathbf{2.61\times}$ \textbf{speed-up in forward passes} with zero additional peak memory overhead. Code is available at \url{https://anonymous.4open.science/r/ENGINE-D0A4}.

Gaze following requires both scene understanding and gaze reasoning to localize the gaze target of an in-scene person. Recently, vision foundation models (VFMs) have demonstrated strong performance on this task, enabling simpler architectures while outperforming prior methods. However, we observe a key limitation of VFM-based approaches: while VFMs substantially improve scene understanding, they contribute little to gaze reasoning. As a result, existing methods often rely on semantically salient objects rather than true gaze cues, leading to degraded performance when targets are not salient. To address this, we propose a novel training mechanism to enhance gaze reasoning in VFMs for gaze following. Our method includes: (1) a head-conditioned local LoRA, which enables localized adaptation to preserve scene token learning while improving head token learning for gaze reasoning; and (2) an out-of-cone penalty, which injects gaze cues into head tokens while aligning them with scene tokens. Experiments on the GazeFollow and VAT datasets demonstrate that our method achieves state-of-the-art performance, with particularly strong improvements when gaze targets are not semantically salient. Our findings offer valuable insights for advancing future gaze following research. We will release the code once the paper is accepted.

Diffusion-based large language models (dLLMs) rely on bidirectional attention, which prevents lossless KV caching and requires a full forward pass at every denoising step. Existing approximate KV caching methods reduce this cost by selectively updating cached states, but their decision overhead scales with context length or model depth. We propose EntropyCache, a training-free KV caching method that uses the maximum entropy of newly decoded token distributions as a constant-cost signal for deciding *when* to recompute. Our design is grounded in two empirical observations: (1) decoded token entropy correlates with KV cache drift, providing a cheap proxy for cache staleness, and (2) feature volatility of decoded tokens persists for multiple steps after unmasking, motivating recomputation of the $k$ most recently decoded tokens. The skip-or-recompute decision requires only $O(V)$ computation per step, independent of context length and model scale. Experiments on LLaDA-8B-Instruct and Dream-7B-Instruct show that EntropyCache achieves 15.2×–26.4× speedup on standard benchmarks and 22.4×–24.1× on chain-of-thought benchmarks against vanilla baselines, with competitive accuracy and decision overhead accounting for only 0.5% of inference time.

Patch-based transformers have emerged as efficient and improved long-horizon modeling architectures for time series modeling. Yet, existing approaches rely on temporally-agnostic patch construction, where arbitrary starting positions and fixed lengths fracture temporal coherence by splitting natural transitions across boundaries. This naive segmentation often disrupts short-term dependencies and weakens representation learning. We propose a novel Entropy-Guided Dynamic Patch Encoder (EntroPE), as a temporally informed framework that dynamically detects transition points via conditional entropy and dynamically places patch boundaries. This preserves temporal structure while retaining the computational benefits of patching. EntroPE consists of two key modules, namely an Entropy-based Dynamic Patcher (EDP) that applies information-theoretic criteria to locate natural temporal shifts and determine patch boundaries, and an Adaptive Patch Encoder (APE) that employs pooling and cross-attention to capture intra-patch dependencies and produce fixed-size latent representations. Extensive experiments on long-term forecasting, classification, and anomaly detection demonstrate that the proposed method improves both accuracy and efficiency, establishing entropy-guided dynamic patching as a promising new paradigm for time series modeling.

We consider multi-environment prediction problems. We assume the environments change the distribution of a latent variable, while the mechanisms generating observed covariates and targets remain stable conditional on that variable. For example, hospitals or clinical cohorts may differ in the prevalence of latent patient states, even though the relationships between those states, physiological measurements, and outcomes remain unchanged. Given a dataset from multiple environments, we formulate a Bayesian model for such problems and derive the corresponding variational objective. We show that this objective decomposes into per-environment terms and an additional cross-environment balancing term induced by the model's structure. We use an empirical Bayes method to set the prior and incorporate it into the objective. Based on this objective, we develop an amortized variational algorithm for posterior approximation, and use the resulting learned latent variables to form predictions in new environments. We study our approach through simulations and real-world studies of astronomical source identification, microbiome-based disease detection, and ICU sepsis prediction. Across these settings, our method outperforms previous approaches for prediction in new environments.

Generative models have made rapid progress in ordered crystal structure prediction, yet many functional materials are intrinsically disordered, with substitutional mixing, vacancies, or interstitial species controlling their properties. Existing crystal generators either assume deterministic site occupations or require site-level disorder annotations, which are often unavailable when the chemical formula is the primary input. We formulate disordered crystal structure prediction through an Occupancy Distribution Matrix (ODM), a continuous site-by-species representation that unifies ordered crystals, solid solutions, vacancy disorder, and interstitial occupancy. A valid ODM must satisfy coupled site-wise occupancy, mass-conservation, and non-negativity constraints, placing each sample on a formula-dependent transportation polytope. We propose Entropic Polytope Flow (EP-Flow), a marginal-constrained flow matching framework that canonicalizes heterogeneous polytopes into a shared double-centered space, learns a marginal-preserving flow, and recovers feasible occupancies through a Sinkhorn inverse map. By jointly generating occupancies, fractional coordinates, and lattice parameters, EP-Flow achieves state-of-the-art performance on formula-conditioned disordered CSP benchmarks derived from COD and MPDS, substantially outperforming adapted ordered-crystal generators. Analyses further show that EP-Flow recovers sparse and chemically meaningful local disorder patterns rather than merely matching global composition statistics.

Video understanding is commonly built on action-triggered responses or scene-level aggregation, implicitly assuming clear temporal boundaries. This assumption breaks in dense-action streams such as gameplay and sports, where overlapping actions induce continuous latent dynamics without explicit transitions. The challenge is further amplified in streaming settings, where models operate under strict causal constraints and must progressively construct episodic memory for downstream reasoning. Existing segmentation-based strategies therefore often produce fragmented or semantically mixed episodes, leading to degraded understanding. We present EpiStream, an online framework for utility-aware semantic episode formation in dense-action streams. Instead of detecting event boundaries, EpiStream learns when an evolving video prefix should be committed into an episode unit that maximizes downstream utility under limited memory budgets. The framework consists of three steps: (i) utility function design for semantic coherence, inter-episode distinctiveness, and downstream objectives; (ii) utility-optimal commit advantage construction from teaching signals; and (iii) peak-aware advantage learning for training a causal online commitment policy from prefix-only observations. Extensive experiments show that EpiStream substantially improves temporal semantic coherence and consistently benefits challenging downstream tasks, including real-time advice and player intention prediction.

We introduce Equilibrium Matching (EqM), a generative modeling framework built from an equilibrium dynamics perspective. EqM discards the non-equilibrium, time-conditional dynamics in traditional diffusion and flow-based generative models and instead learns the equilibrium gradient of an implicit energy landscape. Through this approach, we can adopt an optimization-based sampling process at inference time, where samples are obtained by gradient descent on the learned landscape with adjustable step sizes, adaptive optimizers, and adaptive compute. EqM surpasses diffusion/flow models empirically, achieving an FID of 1.90 on ImageNet 256$\times$256. EqM is also theoretically justified to learn and sample from the data manifold. Beyond generation, EqM is a flexible framework that naturally handles tasks including partially noised image denoising, OOD detection, and image composition. By replacing time-conditional velocities with a unified equilibrium landscape, EqM offers a tighter bridge between flow and energy-based models and a simple route to optimization-driven inference.

Equivariant Graph Neural Networks (GNNs) have significantly advanced the modeling of 3D molecular structure by leveraging group representations. However, their message passing, heavily relying on strictly equivariant operations, suffers from restricted expressiveness due to the limited non-linearity and low degree of group representations. To overcome this, we introduce the Equivariant Spherical Transformer (EST), a novel plug-and-play module that applies a Transformer-based architecture to the Fourier spatial domain of group representations. The integration of EST enhances the model's expressiveness while preserving the crucial equivariant inductive bias through a uniform sampling strategy of spherical Fourier transforms. As demonstrated by our experiments on challenging benchmarks like OC20, MPtrj and QM9, EST-based models achieve state-of-the-art performance. For the complex molecular systems within OC20, small models empowered by EST can outperform some larger models and those using additional data. In addition, we provide both theoretical and experimental validation of EST's equivariance as well, paving the way for new research in this area.

Diffusion Transformers enable powerful in-context image editing from a reference image and a text prompt, but also make it easier to maliciously edit public images while preserving the private identity of the reference. Existing DiT-oriented protections, such as DeContext, mainly suppresses context-to-target attention to weaken reference utilization. However, this strategy can be insufficient for complex edits where prompt semantics already dominate generation. Instead, we study proactive protection for DiT-based editors from a conditional-flow perspective, and use a local geometric interpretation of in-context editing, where the denoising trajectory is empirically characterized as being steered by prompt-consistent and reference-consistent directions. Thus, effective protection should not simply corrupt the output, but should drive it away from the original context manifold. To this end, we propose Conditional Flow Hijacking, which adds imperceptible perturbations to the reference image. Instead of suppressing context attention, our method preserves the context pathway, but redirects the context-induced steering signals in the intermediate transformer layers, causing the final trajectory to drift away from the reference consistent solution region. Experiments on FLUX.1-Kontext with different datasets and prompts including attribute and complex scene editing show that our method significantly suppresses reference identity leakage while maintaining image quality and prompt alignment, outperforming prior protection baselines, especially under complex scene editing prompts. We further verify the robustness of the proposed protection method as well as its transferability across other models, highlighting the need for proactive defenses for in-context generative models.

Reinforcement learning (RL) has become a key training step for improving mathematical reasoning in large language models (LLMs), but it often has high GPU memory usage, which makes it hard to use in settings with limited resources. To reduce these issues, we propose Evolution Strategies with Sharpness-Aware Maximization (ESSAM), a full parameter fine-tuning framework that tightly combines the zeroth-order search in parameter space from Evolution Strategies (ES) with the Sharpness-Aware Maximization (SAM) to improve generalization. We conduct fine-tuning experiments on the mainstream mathematica reasoning task GSM8K. The results show that ESSAM achieves an average accuracy of 78.27\% across all models and its overall performance is comparable to RL methods. It surpasses classic RL algorithm PPO with an accuracy of 77.72\% and is comparable to GRPO with an accuracy of 78.34\%, and even surpassing them on some models. Further generalization experiments show that the models trained with ESSAM exhibit stronger generalization ability. Their average performance achieves the best results on 5 out of 6 datasets, indicating that ESSAM can effectively improve the generalization performance of fine-tuned models. In terms of GPU memory usage, ESSAM reduces the average GPU memory usage by $18\times$ compared to PPO and by $10\times$ compared to GRPO, achieving an extremely low GPU memory usage. In addition, we design an accelerated variant of ESSAM, which achieves nearly a twofold speedup while maintaining the same GPU memory usage as ESSAM, and attains an average accuracy of 78.02\% across all models, outperforming PPO. Code: https://anonymous.4open.science/r/ESSAM-3F4F/

The goal of asymmetric pansharpening is to synthesize a high-resolution multispectral (HR-MS) image by integrating panchromatic (PAN) and low-resolution multispectral (LR-MS) images. While State Space Models (SSMs), particularly Mamba, offer a promising paradigm for global dependency modeling, their standard selective scan mechanisms operate in discrete Cartesian grids, which are ill-suited for distinguishing between structural and spectral features in complex-valued frequency representations. To resolve this, we propose Euler-Mamba, a physics-inspired framework that reformulates the selective scan as a continuous state evolution on polar manifolds. We develop an Eulerian Selective Scan (ESS) mechanism that treats the fusion task as a trajectory of frequency growth. By performing radial scans within the polar domain, ESS governs complex-valued state transitions through the decoupled architecture: the Explicit Phase Rotation (EPR) block facilitates explicit phase rotations for geometric alignment, while the Implicit Magnitude Mapping (IMM) block performs magnitude scaling for spectral calibration. This polar-decoupled evolution allows the model to capture the continuous transition from global spectral backgrounds to fine structural details with linear complexity. Consequently, by modeling the fusion as a continuous functional evolution flow, Euler-Mamba achieves resolution-invariant performance and zero-shot generalization across varying spatial scales, bypassing the computational complexity inherent in traditional discrete-integration frameworks. Extensive experiments demonstrate that Euler-Mamba establishes a new Pareto frontier for asymmetric pansharpening, achieving a superior balance between physical interpretability, computational efficiency, and reconstruction quality.

Frontier large language models can often recognize when they are being evaluated, a capability known as evaluation awareness. If models behave differently in evaluations than in deployment, this undermines the validity of evaluation results, which are a crucial component of current AI safety frameworks. We introduce EvalAwareBench, an open pipeline and benchmark for measuring evaluation awareness that works with any Inspect-compatible evaluation, allowing practitioners to test against current and future benchmarks. EvalAwareBench ships with a newly curated transcript suite covering current frontier system-card evaluations and diverse deployment sources. The benchmark serves two purposes: measuring how reliably frontier LLMs recognize that they are being evaluated, and assessing how detectable individual benchmarks are as evaluations. We identify two methodological choices in the existing literature that introduce systematic bias: the identity of the model that generated the deployment transcripts accounts for 11.25% of measurement variance and can reorder model rankings; and elicitation prompts selected for high performance on one model can perform near chance on others. EvalAwareBench corrects for both via per-model probe calibration and a stratified generator-harmonisation procedure. We release EvalAwareBench as an open pipeline and dataset.

Finetuning Language Models often requires enforcing constraints on individual inputs without compromising performance. However, current alignment methods typically impose constraints only on average, which can induce undesirable disparities across inputs or users. We propose a finetuning framework that addresses this limitation by enforcing alignment requirements as per-sample constraints. To handle the optimization challenges inherent in this approach, we utilize an augmented Lagrangian formulation in the dual domain. Since pointwise constraints can be overly restrictive in low-probability regions or in the presence of outliers, we introduce a learned, sample-dependent relaxation that minimally relaxes constraints to optimize objective performance. We demonstrate the versatility of our framework across three small language model tasks: safety in instruction following, preference satisfaction in function calling, and length-aware re-ranking. Across these settings, our approach reduces tail constraint violations while largely preserving or improving downstream performance

How search strategies evolve in sparse, depletable landscapes remains a central question in foraging theory. We study this problem with an evolutionary simulation in which agents forage on a two-dimensional toroidal lattice containing non-renewable resources distributed either uniformly or as Lévy dust. Each agent carries a heritable genome encoding step lengths, velocities, and turning angles, and selection acts on a fitness function, derived from first principles, combining energetic gain, movement cost, and coverage efficiency. By allowing movement traits to evolve without imposing a prescribed power-law step-length distribution, we test whether selection recovers a strict Lévy-like random walk, similar to the spatial distribution of resources, or instead favors an alternative search mechanism. Strikingly, our results indicate that, in the finite depletion-driven landscapes considered here, evolved search is more consistent with intermittent dynamics than with strict scale-free Lévy motion. To characterize the effective dynamics of the evolved trajectories, we take the average coefficient of determination when fitting second- and fourth-order displacement moments to intermittent search and Lévy-walk models. While a Lévy-like random walk fits very well with the numerical results from the evolutionary search ($R^2>0.9$), the intermittent search achieves a closer fit, with fitted coefficients of determination ($R^2>0.99$) for all resource distributions. Evolution rapidly reshapes the movement genome toward short displacements while retaining a sparse tail of longer relocations, consistent with local exploitation punctuated by occasional transfer. The framework provides a controlled setting for studying how search rules emerge under resource limitation and may inform resource-constrained exploration in autonomous systems.

In this paper, we introduce Exact Flow Linear Attention(EFLA), an exact-flow formulation of delta-rule linear attention. We show that the delta-rule update can be interpreted as an explicit Euler discretization of an underlying continuous-time system. EFLA replaces this first-order update with the exact closed-form flow. By exploiting the rank-1 structure of the dynamics matrix, both the matrix exponential and the input integral collapse to a simple update that preserves delta-rule linear attention's algebraic structure, parameter count, linear-time complexity, and chunkwise parallelism. This attention mechanism removes the Euler discretization error of the delta-rule dynamics without introducing additional parameters. Experiments on robustness tests, language modeling benchmarks, and the MAD synthetic benchmark show that EFLA improves stability under corrupted and high-energy inputs, reduces perplexity, and achieves stronger downstream performance compared to SSM and Euler-style baselines. These results establish exact-flow integration as a principled and scalable update mechanism for delta-rule linear attention.

Flow maps have enabled remarkable progress in few-step generative modeling across both continuous and discrete state spaces. Despite their promise, existing parameterizations are restricted to flows over fixed dimensions or fixed sequence lengths. Here, we introduce Expanding Generative Flows (EFlows), which define flows between distributions of increasing dimensionality through an expanding interpolant built from augmented distributions of conditional noise. Building on this framework, we propose Expanding Flow Maps (EFMs), a new class of flow maps that distill EFlows into efficient few-step generative models. Each EFM factors the map between any two timesteps into two learned components: an expand operator, which augments the state with new coordinates or tokens, and a transport operator, which pushes the expanded state forward along the interpolant. Composing these operators yields a single map that jointly expands and denoises the state, recovering existing fixed-canvas flow maps as the special case in which the expand operator is the identity. We further extend the framework to the discrete simplex, enabling variable-length sequence generation via token insertions. Across domains, EFlows and EFMs provide a principled approach to generative problems in which output size is itself learned, controllable degree of freedom.

Reinforcement learning (RL) has demonstrated success in post-training large language models (LLMs) as agents for tasks such as computer use, tool calling, and coding. However, exploration remains a central challenge in RL for LLM agents, especially as they operate in language-action spaces with complex observations and sparse rewards. In this work, we address exploration for LLM agents by leveraging the ability of LLMs to plan and reason in language about the environment to shift exploration from low-level actions to higher-level language strategies. We thus propose Strategy-Guided Exploration (SGE), which first generates a concise natural-language strategy that describes what to do to make progress toward the goal, and then generates environment actions conditioned on that strategy. By exploring in the space of strategies rather than the space of actions, SGE induces structured and diverse exploration that targets different environment outcomes. To increase strategy diversity during RL, SGE introduces mixed-temperature sampling, which explores diverse strategies in parallel, along with a strategy reflection process that grounds strategy generation on the outcomes of previous strategies in the environment. Across UI interaction, tool-calling, coding, and embodied agent environments, SGE consistently outperforms exploration-focused RL baselines, improving both learning efficiency and final performance. We show that SGE enables the agent to learn to solve tasks too difficult for the base model.

Incorporating diffusion-generated synthetic data into adversarial training (AT) has been shown to substantially improve the training of robust image classifiers. In this work, we extend the role of diffusion models beyond merely generating synthetic data, examining whether their internal representations, which encode meaningful features of the data, can provide additional benefits for robust classifier training. Through systematic experiments, we show that diffusion models offer representations that are both diverse and partially robust, and that explicitly incorporating diffusion representations as an auxiliary learning signal during AT consistently improves robustness across settings. Furthermore, our representation analysis indicates that incorporating diffusion models into AT encourages more disentangled features, while diffusion representations and diffusion-generated synthetic data play complementary roles in shaping representations. Experiments on CIFAR-10, CIFAR-100, and ImageNet validate these findings, demonstrating the effectiveness of jointly leveraging diffusion representations and synthetic data within AT.

SHAP explanations are widely used in high-stakes settings to justify decisions, yet they can differ substantially across repeated runs, even when the model, the input instance, and the prediction are held fixed. Prior work has documented disagreement *between* explanation methods; we show that substantial disagreement arises even *within* SHAP across reruns of the same estimator on the same trained model and instance. We call this phenomenon *explanation multiplicity* and develop an evaluation methodology for characterizing it under deployment-realistic computational budgets, combining a dual-seed protocol that disentangles model-induced from explainer-induced variability, a hierarchy of magnitude-, rank-, and set-based metrics, and randomized Dirichlet and Mallows null models that calibrate when observed disagreement exceeds chance. Across multiple datasets, model classes, and sampling strategies, we find that explanation multiplicity is pervasive and persists even for high-confidence predictions. The dominant source depends on the data regime: model-induced variability dominates in small-data settings, while explainer-induced variability dominates at scale. Commonly used $\ell_2$ distance is largely insensitive to this instability, while rank-based metrics reveal severe reordering of top-ranked features. Improved sampling methods such as CTE do not eliminate rank-level multiplicity, and deterministic alternatives such as K-Means achieve stability only by targeting a different estimand that diverges sharply from Shapley values defined against the empirical training distribution. Practitioners should treat single-run SHAP outputs as realizations of a distribution rather than as authoritative artifacts.

Vision-language-action(VLA) models have emerged as a promising interface for autonomous driving. However, existing VLA models still suffer from a fundamental mismatch: driving actions require precise 3D geometric cues, while visual-language understanding and reasoning are largely conducted in a 2D semantic space. In this paper, we propose GeoCoTDrive, an explicit geometric chain-of-thought framework that grounds geometry in a planning-oriented manner. GeoCoTDrive follows a ``think with 2D first, drive with dedicated 3D priors'' paradigm: it first identifies sparse 2D regions corresponding to decision-critical cues, and then retrieves localized 3D priors by sampling features from a geometric foundation model within the grounded regions. These localized geometric features are interleaved into the autoregressive context to support trajectory generation. To supervise this process, we introduce planning-relevant grounding, a new region-level grounding task that focuses on local spatial cues directly affecting ego planning, and construct the PlanningGrounding dataset to endow VLM models with planning-oriented grounding ability. Experiments across multiple end-to-end autonomous driving benchmarks show that GeoCoTDrive consistently improves safety-critical planning performance, demonstrating the effectiveness of explicit geometric chain-of-thought reasoning for VLA-based planning.

Class-wise embeddings provide an effective strategy for multi-label learning (MLL) by capturing the distinct discriminative properties of each class. However, existing methods based on class-wise embeddings typically overlook the internal distribution of negative samples. We uncover a pivotal phenomenon: negative instances in class-wise spaces naturally organize into multiple semantically coherent sub-clusters, a structure that persists across diverse modalities even under extreme label sparsity. Leveraging this insight, we propose a Structure-Aware Pseudo-label mining framework for weakly supervised MLL (SaPu). The proposed framework mines positive pseudo-labels by aggregating cluster-level consistency across different class-wise embeddings and identifies negative pseudo-labels via neighborhood exclusion. These high-quality signals further guide class-wise contrastive learning, establishing a self-reinforcing loop that iteratively refines embeddings. Extensive experiments on image, text and audio benchmarks validate that SaPu outperforms state-of-the-art methods by an average of 4.46% in challenging single-label supervision scenarios. Code is available in the supplementary material.

Vision Language Action (VLA) models can execute isolated atomic skills, but practical long horizon manipulation requires recombining these skills into new compositions without demonstrations for every skill combination. The direct full instruction route treats skill chaining as long horizon imitation. In our controlled study, full instruction training on limited skill combination sequences reaches 40.0% ordered success on covered combinations but only 7.7% on recombination, suggesting memorization of sequence templates rather than robust skill recombination. We therefore introduce Hierarchical Subtask Execution (HSE), a diagnostic execution framework that exposes skill recombination through a shared modular interface over the planner, trigger, and policy, enabling long horizon failures to be decomposed into separable modes. To make this diagnosis measurable, we introduce RoboCombine, a benchmark for skill recombination without combined skill training trajectories, with metrics for ordered success and transition conversion across subtask boundaries. RoboCombine reveals that atomic competence alone does not ensure ordered execution. Failures concentrate at transitions where downstream subtasks must launch from handoff states produced by preceding subtasks rather than canonical atomic starts. We identify this mismatch as the start state gap. To address it, we introduce Pose Robust Task Initialization (PRTI), a lightweight atomic task post training strategy that broadens start state coverage beyond canonical starts. Under oracle timed HSE on RoboCombine, adding PRTI raises ordered success from 2.11% to 26.32% and transition conversion from 2.63% to 60.53%, without combined skill training trajectories. Together, these results show that combinatorial generalization requires handoff robust atomic skills that convert first stage progress into reliable downstream execution, not merely decomposition into known subtasks. We will release RoboCombine and HSE code to enable reproducible diagnosis of atomic skill recombination failures.

Identifying latent dynamical systems from noisy, high-dimensional measurements is a central problem at the intersection of representation learning, system identification, and scientific discovery. We present DYSCO, a multi-view temporal contrastive learning algorithm that jointly recovers latent trajectories and the governing dynamics from such observations, by leveraging multiple independent noisy views of the same underlying process to disentangle signal from noise. By parameterizing the dynamics in a structured functional basis, our framework further enables symbolic recovery of the governing equations within an affine gauge. We offer theoretical guarantees for strong identification up to an affine indeterminacy, extending prior identifiability results to the realistic setting of noisy nonlinear observations. Empirically, we demonstrate accurate recovery of both latent trajectories and flow fields across a diverse set of dynamical regimes (e.g, chaotic, oscillatory, and metastable) under both Gaussian and Poisson observation noise, the latter being particularly relevant for neural recordings.

Cognitive Load Theory (CLT) shows that human problem-solving is affected by both intrinsic load, arising from task difficulty, and extraneous load, arising from how task information is presented. LLMs show analogous effects, where presentation alone can affect model performance, but these effects have not yet been systematically isolated from task difficulty. To address this gap, we introduce CogLoadBench, a controlled benchmark that varies five CLT-inspired presentation factors across three reasoning task families. Across models, we find that higher extraneous load conditions often reduce accuracy by either adding competition or complicating the reasoning path. To support diagnosis and mitigation of load-related effects, we propose the Extraneous Load Score (ELS), a prompt-time metric computed from activations that estimates a model's extraneous load, and show that it generalizes across reasoning task types and load levels. We further show that shifting model representations toward lower-ELS directions at inference time can significantly improve model performance without changing prompt content.

This paper studies category-level RGB-D object pose estimation, which recovers an object's 3D rotation and translation without instance-specific CAD models or reference views. This task is highly ambiguous due to intra-class variations, symmetries, and partial observations. Current solutions struggle to balance reliability and efficiency: deterministic methods are prone to error propagation from intermediate structures, while diffusion models suffer from high inference latency and complex post-hoc candidate extraction. To address this, we present F2G-Pose, a real-time direct, single-pass framework that predicts category-level pose, metric size, and camera-frame completed shape from a single RGB-D observation in one feed-forward pass. F2G-Pose lifts dense visual foundation features to partial point clouds, encodes local geometry with DGCNN tokenization, and integrates object-level structure with a Mixture-of-Experts Transformer. Joint size prediction and camera-frame shape completion encourage metric consistency through dense geometric supervision. Trained only on synthetic SOPE data, F2G-Pose achieves state-of-the-art accuracy on SOPE and ROPE while running at 28 FPS, and further demonstrates strong camera-frame geometry on HouseCat6D and real-object transfer on HANDAL. Analyses of latency, robustness, stability, shape completion, and GenPose++ candidate behavior further demonstrate the efficiency and reliability of F2G-Pose, with real-world manipulation experiments demonstrating deployment potential under segmented RGB-D inputs.

Vision-language-action (VLA) models provide a promising paradigm for scalable robotic manipulation, yet their reliance on success-only behavioral cloning leaves them brittle; lacking corrective training signals, minor execution errors rapidly compound into unrecoverable, out-of-distribution failures. To address this limitation, we propose Adaptive Failure-Informed Learning (AFIL), an end-to-end framework that leverages failure trajectories as adaptive negative guidance for diffusion- and flow-based VLA policies. AFIL uses a pretrained VLA to generate failure rollouts online, avoiding the need for handcrafted failure-mode design or human-in-the-loop recovery. It then jointly trains Dual Action Generators (DAGs) for successful and failed behaviors while sharing a common vision-language backbone, enabling efficient failure-aware policy learning with limited parameter overhead. During sampling, the failure generator adaptively steers action generation away from failure-prone regions and toward more reliable success modes, with guidance strength determined by the per-diffusion-step distance between success and failure distributions. Experiments across in-domain and out-of-domain robotic manipulation tasks, covering both short- and long-horizon settings, show that AFIL consistently improves task success rates and robustness over existing VLA baselines, demonstrating its effectiveness, efficiency, and generality.

Filtered vector search in private and on-prem retrieval-augmented generation (RAG) systems must control tail risk under fixed latency, RAM, and SSD budgets, not merely pick a fast backend. Our measurements isolate a recurring ambiguity regime concentrated in the 5-15% failure band, where small shifts in vector-metadata alignment flip which familiar P1-P4 action is safe and create 3-6 P95 spread. We study that regime as authorization over fixed actions rather than as generic planning. The paper contributes a measured failure-band stress test, an auditable authorize/veto rule over fixed P1-P4 actions, and PACER as one concrete controller that combines recall-floor adjustment, calibrated tail intervals, and regime-local fallback. The measured failure band exposes a different runtime decision boundary: when the provisional winner is separated enough to trust and when uncertainty should refuse it. On Medium+EntDoc, evaluated on a dual-AMD EPYC 7543 server with 128GB RAM at matched recall in the 5-15% failure band, PACER reduces P95 by 46%\ and P99 by 45%\ relative to the strongest baseline while maintaining 0.94-0.98\ Recall@20. The gain concentrates where plan rankings are unstable and weakens when the authorization logic is stripped back, while feasibility, transfer, drift, and hierarchy-aligned slices remain supporting diagnostics. For recurring fixed-budget filtered retrieval, these results show that an auditable authorize/veto layer over fixed actions can materially stabilize tail latency in the measured failure band.

Training smaller language models on reasoning solutions generated by a larger teacher model is the prevailing approach for transferring mathematical reasoning capabilities. However, the student tends to overfit the output distribution of the teacher and generalizes poorly when its own behavior at inference time deviates from the training data. Recent methods attempt to bridge this gap by incorporating error data, yet the corrective trajectories still reflect the reasoning patterns of the teacher, which the student cannot reliably reproduce, leaving the distribution mismatch largely unresolved. We propose a method built on the insight that effective training data should be jointly determined by what the student fails on and what it can realistically learn. Our method first profiles the failure behavior of the student on each problem through repeated sampling, characterizing it in terms of error rate and error consistency to reveal qualitatively distinct failure modes. Each mode demands a fundamentally different training signal. The diagnosed mode then guides candidate generation from the teacher via distinct prompting strategies tailored to each failure type. Among the correct candidates, our method selects the one whose initial reasoning steps after the point of divergence are most reproducible by the student. These first corrective steps constitute the primary bottleneck, while the remaining steps follow an already corrected context and can therefore be continued more readily. This contrasts with standard reranking approaches, which score the trajectory as a whole and thereby dilute the signal at the critical reasoning transition. Experiments on GSM8K, MATH, and three out-of-distribution benchmarks for Llama-3.1-8B and DeepSeek-Math-7B demonstrate that our method achieves superior performance compared with other strong methods. The code and trained weights will be publicly available.

Autoregressive perception models trained to localize visual entities under the open-vocabulary setting are mostly trained using Supervised fine-tuning (SFT) with maximum likelihood, yet it optimizes a proxy objective (per-token cross-entropy) that is fundamentally misaligned with perception metrics such as precision and recall. In this paper, we explore post-training reinforcement learning (RL), specifically GRPO, to directly align these models with their evaluation metrics. Building up on the recently introduced Falcon Perception, we design an RL framework that addresses perception-specific challenges: reward design for set-structured outputs and multi-head sampling control. We discover multiple benefits from RL for perception: first, RL unlocks unprecedented performance in very dense scenes (up to 500 objects per scene) and to the best of our knowledge, no other system has such capabilities at this scale; furthermore it fixes common issues in autoregressive perception models like mask repetitions and removes almost entirely the need for NMS and coordinate deduplication, which improve both performance and efficiency and remove the need for hyperparameters tuning; overall, we notice improvements on all levels of difficulties in referring expression segmentation (on PBench and SACO-Gold), and we find an elegant way to preserve the knowledge of whether an object exists or not (as evaluated by MCC) without training on negative samples. We show that a simple reward penalizing false negatives and positives is enough. We develop two hybrid self-annotation pipelines, respectively tailored for difficult referring expressions and very dense scenes, and show their benefits on RL-training. Model weights and datasets will be made public.

4D mesh generation has recently emerged as a powerful paradigm for recovering dynamic 3D structure from videos, but existing methods remain slow, computationally expensive, and difficult to scale to longer sequences. We introduce a training-free approach that accelerates 4D mesh generation while improving temporal correspondence quality. Our key observation is that temporal correspondences emerge inside a 4D backbone long before its generated meshes become visually accurate. We exploit this with a general framework we call Spatio-Temporal Attention Chain which propagates information across space and time. Starting from vertices on an anchor mesh, the chain maps vertices to latent tokens. It then follows temporal correspondences in latent space, and recovers frame-specific vertices through latent-to-vertex attention. This design avoids expensive explicit matching while preserving anchor mesh details and thereby improving dynamic mesh geometry and temporal consistency. Compared to state-of-the-art, our method generates a 4D mesh in 9 seconds, achieving a $13\times$ speedup while producing higher-quality results. Moreover, our approach scales to videos up to $16\times$ longer without degrading mesh quality. Beyond generation, the improved correspondences enable competitive zero-shot performance on two downstream tasks: 2D object tracking and 4D tracking. We further show that our framework enables reliable camera estimation, a capability not supported by prior 4D mesh generation methods.

Diverse applications on signed networks, ranging from leader-follower opinion dynamics to graph semi-supervised learning, fundamentally reduce to solving the discrete Dirichlet boundary value problem. The computational bottleneck of this problem lies in evaluating the entries of the label propagation operator $Q = -L_{U,U}^{-1}L_{U,L}$. While existing approximation methods based on absorbing random walks attempt to address this complexity, they often suffer from high variance and limited precision. To overcome these limitations, we propose MultiTreeQ, a novel and efficient framework designed to approximate the operator's entries with high accuracy. Specifically, MultiTreeQ is constructed based on the Signed Multi-Rooted Spanning Tree and leverages Rao-Blackwellized variance reduction to significantly improve estimation precision. Extensive experiments on real-world networks confirm that MultiTreeQ substantially outperforms existing baselines, scaling to massive graphs with up to 10 million nodes while delivering remarkable improvements in computational efficiency and precision. Our code is available at https://anonymous.4open.science/r/label-propagation-operator-2059/.

Structure learning for sparse graphical models in high dimensions requires both computational scalability and selection consistency. Classical criteria such as (Extended) Bayesian Information Criterion can provide consistency, but rely on model-specific likelihood and complexity calibration. In comparison, cross-validation avoids such analytic calibration and is therefore attractive for graph learning. However, cross-validation is computationally infeasible and theoretically inconsistent in structure learning. In this work, we resolve the tension by developing a scalable graph-recovery procedure built on approximate cross-validation. We demonstrate its effectiveness in learning large graphical models both theoretically and empirically, and establish selection consistency under mild signal strength assumptions.

Modeling resting-state functional magnetic resonance imaging (rs-fMRI) data is crucial for understanding brain-wide neural activity. However, traditional methods struggle to capture complex temporal dynamics over long horizons, to account for the brain's anatomical spatial structure, and to model high-dimensional ambient signals that lie on a low-dimensional intrinsic subspace. We propose FAST-Brain, a unified flow-aligned spatio-temporal surrogate brain model that addresses all three challenges. At its core is a flow-aligned generative framework that directly predicts the clean blood-oxygen-level-dependent (BOLD) signal, paired with a graph convolutional network that captures spatial structural constraints and a Transformer that models long-range temporal dependencies. Theoretically, we show that under a low-dimensional subspace assumption, the approximation error of our model scales with the intrinsic dimension rather than the ambient dimension, which justifies our direct modeling of the BOLD signal. Extensive experiments on synthetic and Human Connectome Project datasets demonstrate that FAST-Brain achieves state-of-the-art performance in recovering functional connectivity, effective connectivity, and the implicit low-dimensional signal subspace.

We study clustering algorithm for dynamic evolving graphs $\{ G_t\}$, in which new edges (and potentially new vertices) are added into a graph, and the underlying cluster structure of the graph gradually changes over time. By designing a hierarchical data structure for dynamically maintaining the cluster structure of $\{ G_t\}$, we prove that the cluster structure of $G_T$ can be maintained with $O(d)$ amortised update time and $\widetilde{O}\left(n_T^{1/d}\right)$ amortised query time, for an arbitrary large constant $d$. This significantly improves the previous state-of-the-art (Laenen and Sun, ICML~'24), which requires $O(1)$ update time and $O(n_T/\log n_T)$ amortised query time. We further demonstrate this improvement with experiments on both synthetic and real-world datasets.

Efficient key-value (KV) cache management is crucial for the practical deployment of large language models (LLMs), yet existing compression techniques often incur a trade-off between performance degradation and computational overhead. We propose a novel gating-based KV cache eviction method for frozen-weight LLMs that achieves high compression ratios with negligible computational cost. Our approach introduces lightweight sink-attention gating modules to identify and retain critical KV pairs, and integrates seamlessly into both the prefill and decoding stages. The proposed gate training algorithm relies on forward passes of an LLM, avoiding expensive backpropagation, while achieving strong task generalization through a task-agnostic reconstruction objective. Extensive experiments across the Qwen2.5-1M, Qwen3, and Gemma3 families show that our method maintains near-lossless performance while evicting up to 70\% of the KV cache. The results are consistent across a wide range of tasks, including long-context understanding, code comprehension, and mathematical reasoning, demonstrating the generality of our approach.

World Action Models (WAMs) have emerged as a promising alternative to Vision-Language-Action (VLA) models for embodied control because they explicitly model how visual observations may evolve under action. Most existing WAMs follow an imagine-then-execute paradigm, incurring substantial test-time latency from iterative video denoising, yet it remains unclear whether explicit future imagination is actually necessary for strong action performance. In this paper, we ask whether WAMs need explicit future imagination at test time, or whether their benefit comes primarily from video modeling during training. We disentangle the role of video modeling during training from explicit future generation during inference by proposing **Fast-WAM**, a WAM architecture that retains video co-training during training but skips future prediction at test time. We further instantiate several Fast-WAM variants to enable a controlled comparison of these two factors. Across these variants, we find that Fast-WAM remains competitive with imagine-then-execute variants, while removing video co-training causes a much larger performance drop. Empirically, Fast-WAM achieves competitive results with state-of-the-art methods both on simulation benchmarks (LIBERO and RoboTwin) and real-world tasks, without embodied pretraining. It runs in real time with 110 ms latency, over 6$\times$ faster than existing imagine-then-execute WAMs. These results suggest that the main value of video prediction in WAMs may lie in improving world representations during training rather than generating future observations at test time.

Federated unlearning aims to efficiently remove the influence of a specific client from a trained global model in federated learning without full retraining. However, existing approaches struggle to preserve the utility of the remaining clients because unlearning can be unstable at both the client and server stages. Local forgetting updates may become overly aggressive and destabilize client side optimization, while server aggregation can amplify conflicts between forgetting updates and retained client updates. To address both failure modes within a unified framework, we propose Federated Unlearning with Gradient Adaptive Shaping (FUGAS), a history-free gradient shaping framework that stabilizes the entire unlearning pipeline. On the unlearning client side, we employ a bounded preference objective that utilizes the pre-unlearning model predictions on the forgetting data as negative references to controllably steer the model away from unlearning client knowledge while requiring only minimal storage for reference caches rather than full historical updates. On the server side, we introduce a compatibility projection mechanism that reshapes the aggregated unlearning update to remain compatible with directions estimated from retained clients. We provide a theoretical analysis indicating that FUGAS promotes stability and ensures non-increasing empirical risk on retained distributions while establishing an excess risk bound relative to retraining. Extensive experiments demonstrate that FUGAS achieves effective unlearning while consistently maintaining high accuracy on retained data.

Federated adaptation of large vision-language models (VLMs) is appealing for privacy-sensitive applications but must contend with client data heterogeneity, strict communication budgets, and on-device inference constraints. Recent methods predominantly introduce additive adaptation interfaces, such as learnable prompts or adapter modules, which attach extra trainable components to the frozen backbone, inflating both parameter counts and inference latency. We challenge this additive paradigm and ask: can robust federated VLM adaptation be achieved through a compact intrinsic interface alone? We answer this question with \textbf{FedIBS} (Federated Intrinsic Bias Selection), a module-free framework that fine-tunes only the biases in feed-forward network projections. FedIBS operates on existing backbone parameters and introduces zero additional inference cost. To make such compact intrinsic adaptation viable under heterogeneous clients, we further propose a Fisher-guided progressive masking mechanism that disentangles shareable and personalized bias coordinates. Each client retains personalized coordinates locally and communicates only the shareable subset for aggregation, enabling sparse communication while preserving client-specific adaptation. Extensive experiments on diverse datasets and heterogeneity settings demonstrate that FedIBS consistently outperforms recent federated prompt- and adapter-based baselines in both personalization and generalization, all while achieving substantially lower communication cost.

As privacy, copyright, and safety requirements evolve, federated learning systems face a growing imperative to accommodate sequential unlearning requests. However, existing federated unlearning methods are designed for single-shot settings and their direct sequential application may lead to uncontrolled parameter-space overlap, causing cross-request interference that undermines prior forgetting. To address this, we propose the Spectral Orthogonality for Federated Continual unlearning (FedSOUL), a framework that reduces direct overlap among sequential unlearning updates. The core idea is to structure model updates in a shared orthonormal spectral basis, where each client is assigned a dedicated subspace in the frequency domain. Under this design, each unlearning request updates only its associated subspace, while all others remain unchanged, ensuring that sequential updates do not overlap in the allocated spectral coefficient space and thereby reducing cross-request interference. Beyond interference control, FedSOUL also improves efficiency by operating on sparse spectral coefficients, using FFT-based transforms for efficient computation and communicating only sparse coefficient updates. From a theoretical perspective, we relate continual unlearning to geometric conditions on update directions and show that our spectral construction satisfies update orthogonality by design, while functional preservation depends on bounded projected gradient leakage. Extensive experiments demonstrate that FedSOUL maintains stable unlearning–utility trade-offs across up to ten sequential requests and consistently outperforms competitive federated baselines. Our code is available at https://anonymous.4open.science/r/fedunlearning-1B92/.

Feed-forward novel view synthesis has recently shown promising results from sparse posed images, but most existing methods assume that context and target views share a fixed camera family. This homogeneous-camera assumption breaks in practical multi-sensor systems, where perspective, fisheye, and panoramic cameras may coexist and where the target projection may be unseen during training. We study feed-forward NVS across heterogeneous central cameras and identify a key ambiguity introduced by tokenization: a visual token aggregates a projection-dependent bundle of pixel rays, while existing camera encodings mainly expose absolute rays or token-center relations. To address this, we combine token-center relative Camera Positional Encodings and proposed local raymaps, a token-level representation that explicitly describes the intra-patch ray distribution summarized by each token. We further propose projection-aware 2D RoPE, which replaces raw image-grid coordinates with ray-induced angular coordinates so that relative positional reasoning is aligned across camera projections. Together, these components treat diverse cameras as calibrated samplings of a shared ray space rather than separate visual domains. On ScanNet++ with heterogeneous-camera system, our method improves over camera-conditioned baselines under mixed-camera evaluation and demonstrates zero-shot generalization to panoramic views.

Detecting abnormal changes in server operational data is an important task in modern computing systems, as unexpected shifts in traffic, workload, or performance metrics may indicate service degradation, abnormal access patterns, system failures, or potential security risks. However, server monitoring data are often naturally observed as functional time series, with complex temporal patterns, high dimensionality, and heterogeneous structures, making traditional change-point detection methods difficult to apply reliably in practice. To address this problem, we propose a general change-point detection framework based on a Functional Embedding Neural Network (FENet), designed to provide robust detection across different data types and structural conditions without relying on strong prior assumptions. The proposed method is motivated by the connection between the classical functional CUSUM (FCUSUM) statistic and neural network representations, which allows FENet to learn effective detection rules directly from data while retaining useful statistical intuition. We provide theoretical analysis to formalize this connection and study key properties of the proposed approach. Extensive simulation studies show that FENet achieves competitive performance compared with existing methods and remains robust under limited training data and distributional mismatch between training and test settings. We further apply FENet to real-world server operational metrics, demonstrating its practical value in detecting and localizing abnormal changes in complex server monitoring systems.

Humans can learn new tasks from just a few demonstrations. Prior methods like behavior cloning struggle to replicate this ability because they imitate demonstrations without learning the underlying concepts, such as goals, positional relations, and constraints. To address this challenge, we propose COIN, an approach that represents behavior as a composition of reusable concepts. A new task is learned by inferring its concepts from demonstrations then using a policy to implement them. We show that this approach is effective at few-shot learning across diverse domains. On object rearrangement, goal-oriented navigation, and human motion generation, COIN recombines previously seen concepts to solve novel tasks at inference time. On the LIBERO robotics benchmark, COIN achieves a $96.9$% average success rate on LIBERO-90, surpassing pretrained policies such as $\pi_0$ with roughly $10\times$ fewer parameters. This performance gap widens further when fine-tuning on unseen tasks with limited demonstrations, highlighting the advantage of concept inference over end-to-end policy adaptation in the few-shot regime.

Image steering aims to provide fine-grained control over concepts in generated images, enabling targeted and continuous manipulation without broadly altering the generation process. Recent steering methods for diffusion models primarily focus on the text-conditioning space or rely on costly auxiliary training. Deviating from that, we here propose a novel steering approach that extracts visual concept directions directly from the image-side activations of Diffusion Transformers. Specifically, we showcase few-shot concept extraction in the activation space using a simple difference-of-means estimator over intermediate residuals. We observe (a) scene-level concepts such as lighting, weather, and style are distributed across image tokens, as well as (b) local concepts such as object colors, textures, materials, and facial attributes are concentrated in specific regions. Towards controlling both concepts, we introduce a new unified training-free framework that uses global difference-of-means for (a) distributed concepts and masked difference-of-means for (b) local concepts, preventing local signals from being diluted by global averaging. Notably, using only a handful of positive and negative prompt pairs, our method identifies and extracts reusable concept directions that can be injected at inference for continuous steering. Quantitatively, our method outperforms text-encoder steering and training-free activation-editing baselines, as well as is competitive with specialized editing models. Our results suggest that as concept directions can be applied at patch level, our method enables precise targeted and compositional steering, manipulating multiple concepts in the same image.

All-atom generative modeling of $3\mathrm{D}$ biomolecular complexes has emerged as the dominant paradigm for predicting the structure of proteins and protein-ligand systems. Generating structures at the atomic level of fidelity, however, typically requires expensive iterative diffusion rollouts, making both conventional deployment and inference-time search techniques computationally costly. In this paper, we introduce the $\textbf{De}$noiser $\textbf{C}$ofolding $\textbf{A}$ll-atom $\textbf{F}$lowmap (DeCAF) framework for distilling state-of-the-art all-atom cofolding models into all-atom flow maps that produce high-quality samples in only a few inference steps. We build DeCAF on a denoiser-based formulation of flow maps with endpoint losses that naturally support $\mathrm{SE(3)}$ rigid alignment, which we show is critical for training accurate models. We further derive a simple change of variables that lets DeCAF operate in the $\sigma$-space noise schedule of EDM-style architectures, enabling direct distillation from pretrained cofolding diffusion models. Equipped with DeCAF's flowmap lookahead, we introduce a purpose-built inference-time framework that improves sampling through reward-guided search. Empirically, DeCAF statistically improves over Boltz-1(x) in both accuracy (RMSD) and physical validity scores of protein-ligand poses at strict NFE budgets on the challenging Runs N’ Poses, while also showing a more optimal Pareto frontier across all inference compute budgets on PoseBusters.

Multimodal foundation models integrate heterogeneous signals across modalities, yet it remains poorly understood how their predictions depend on specific internal feature groups and whether such reliance can be deliberately controlled. Existing studies of shortcut and spurious behavior largely rely on post hoc analyses or feature removal, offering limited insight into whether reliance can be modulated without altering task semantics. We introduce FiLoRA (Focus-and-Ignore LoRA), an instruction-conditioned, parameter-efficient adaptation framework that enables explicit control over internal feature reliance while keeping the predictive objective fixed. FiLoRA decomposes adaptation into feature group-aligned LoRA modules and applies instruction-conditioned gating, allowing natural language instructions to act as computation-level control signals rather than task redefinitions. Across text-image and audio-visual benchmarks, we show that instruction-conditioned gating induces consistent and causal shifts in internal computation, selectively amplifying or suppressing core and spurious feature groups without modifying the label space or training objective. Further analyses demonstrate that FiLoRA yields improved robustness under spurious feature interventions, revealing a principled mechanism to regulate reliance beyond correlation-driven learning.

Evaluating true metacognition in Large Language Models (LLMs) is difficult due to biases and heuristics. This paper presents a framework to measure and enhance LLM metacognition while controlling for these biases. A measurement method using the $d_{\rm{type2}}'$ metric is systematized to isolate metacognitive ability. The Evolution Strategy for Metacognitive Alignment (ESMA) is proposed, demonstrating robust generalization across unseen datasets, languages, and newly acquired knowledge. Finally, parameter analysis reveals that these improvements are driven by a sparse set of parameters, offering new pathways for targeted metacognitive optimization.

The financial domain poses substantial challenges for Vision-Language Models (VLMs) due to specialized chart formats and knowledge-intensive reasoning requirements. However, existing financial benchmarks are largely single-turn and rely on limited question formats, which hinders comprehensive evaluation in realistic application scenarios. To address this gap, we propose FinMTM, a multi-turn multimodal benchmark that expands diversity along both data and task dimensions. On the data side, we curate and annotate 11,133 bilingual (Chinese and English) financial QA pairs grounded in commonly used financial visuals, including candlestick charts, statistical plots, and report figures. On the task side, FinMTM covers single and multiple choice questions, multi-turn open-ended dialogues, and agent-based tasks. We further design task-specific evaluation protocols, including a set-overlap scoring rule for multiple choice questions, a weighted combination of turn-level and session-level scores for multi-turn dialogues, and a composite metric that integrates planning quality with final outcomes for agent tasks. Extensive experimental evaluation of 22 VLMs reveal their limitations in fine-grained visual perception, long-context reasoning, and complex agent workflows.

Modern models can have high average accuracy and benign loss tails while remaining fragile on rare subpopulations. We identify an information-geometric mechanism: after nuisance projection, the task direction can become locally nonidentifiable even when loss and raw Fisher look well behaved. Fisher-Glass certifies this failure by applying CVaR to inverse nuisance-projected task information. We prove a loss--information separation theorem and show that, under a boundary-mass condition, the hidden-environment tail of inverse projected Fisher controls robust classification sample complexity. We then derive stable ridge certificates and repair principles: reserves lift Fisher-null tails, portfolios increase collapse codimension, and tail-transverse influence selects counterfactual repairs by combining Fisher-Glass directionality with loss leverage. Synthetic experiments validate the theory; WILDS/Waterbirds diagnostics show that weakest identifiability need not mean lowest accuracy; and Waterbirds training plus counterfactual repair show that Fisher-Glass controls an identifiability axis complementary to loss.

Looped-in-depth architectures provide an inductive bias toward learning step-by-step procedures for tasks that require compositional reasoning. The effective depth reached by looping determines the quality of the solution. Similar to deep architectures, looped architectures are prone to signal propagation issues as the halting decision is postponed. In this paper, we address these signal propagation issues by using pre-norm layers and residual scaling. Furthermore, we propose FPRM: a Fixed-Point Reasoning Model that uses fixed-point convergence as an end-to-end halting mechanism in a looped architecture. We show that fixed-point halting allows FPRM to adapt its compute to the difficulty of the task. FPRM proves effective on common reasoning benchmarks, namely sudoku, maze, and state tracking.

In this work, we introduce FLAME, a family of extremely lightweight and capable Time Series Foundation Models, which support versatile forecasting tasks via generative probabilistic modeling, while ensuring both efficiency and robustness. FLAME utilizes the Legendre Memory for strong generalization capabilities. Through adapting variants of Legendre Memory, i.e., translated Legendre (LegT) and scaled Legendre (LegS), in the Encoding and Decoding phases, FLAME can effectively capture the inherent inductive bias within data and make efficient long-range inferences. To enhance the accuracy of probabilistic forecasting while keeping efficient, FLAME adopts a Normalizing Flow based forecasting head, which can model the arbitrarily intricate distributions over the forecasting horizon in a generative manner. Comprehensive experiments on four well-recognized benchmarks, including TSFM-Bench, ProbTS, TFB, and GIFT-EVAL, demonstrate that FLAME is a strong out-of-the-box tool of decision intelligence.

As video becomes increasingly central to information dissemination and multimodal large language models (MLLMs) continue to advance, evaluating video retrieval has become increasingly important. In realistic search scenarios, this requires matching short user queries to long-form content using both visual and auditory evidence. Yet existing retrieval benchmarks are still dominated by short clips, single modalities, and caption-based evaluation. We introduce FLARE, a full-modality long-video audiovisual retrieval benchmark with user-simulated queries. Built from 399 carefully screened Video-MME videos (10--60\,min, 225.4\,h) to ensure source quality and diversity, FLARE contains 87,697 clips annotated with vision, audio, and unified audiovisual captions, together with 274,933 user-style queries. Cross-modal queries are further filtered by a hard bimodal constraint, requiring retrieval to fail under either modality alone but succeed when both are combined. FLARE evaluates models under two regimes, caption-based and query-based retrieval, across vision, audio, and unified audiovisual settings. Experiments with 15 representative retrievers show that user-style queries substantially change model behavior, strong caption-based performance does not always transfer to query-based retrieval, and audio--language alignment remains a key bottleneck for unified audiovisual retrieval. Our code and data are released at https://anonymous.4open.science/r/FLARE-950E/ and https://huggingface.co/datasets/AnonymousFLARE/FLARE

State-space models (SSMs) face a fundamental trade-off between efficiency and expressivity that is mainly dictated by the structure of the model's transition matrix. Unstructured transition matrices enable maximal expressivity, as measured by their ability to model finite-state automaton (FSA) transitions, but come at a prohibitively high compute and memory cost. In contrast, diagonal and other structured transition matrices suffer from limited expressivity, but are highly efficient both in runtime and memory consumption. Building on recent work on structured sparse SSMs, we propose Flash PD-SSM, a novel SSM that achieves comparable throughput to diagonal models with optimal expressivity guarantees. Flash PD-SSM keeps a collection of structured sparse matrices, from which a single matrix is selected at each time-step, enabling FSA expressiveness at the level of unstructured matrices while maintaining the efficiency required for training models at scale. First, we validate Flash PD-SSM against a suite of alternative models on common mechanistic and synthetic state-tracking tasks, showing that its theoretical expressivity is achieved in practice. Moreover, on multivariate time-series tasks involving sequences of length over 17,000, Flash PD-SSM defines a new state-of-the-art (SoTA) accuracy among competing SSM methods. Second, we demonstrate that Flash PD-SSM is an effective drop-in replacement for hybrid LLMs, yielding improvements both in natural language state-tracking and in common language modeling scenarios. Finally, we show that our highly efficient design results in increased throughput and decreased memory consumption with respect to SoTA SSMs widely used in frontier language models.

Diffusion and flow matching have emerged as expressive generative planners for autonomous driving planning, owing to their ability to model high-fidelity and multi-modal trajectory distributions. Nevertheless, existing generative planners are predominantly optimized through imitation learning, which induces a fundamental mismatch between supervised trajectory fitting and closed-loop planning metrics, while their iterative sampling procedures often impose substantial computational overhead. In this paper, we propose FlashPlanner, a goal-conditioned flow-matching planner with online RL finetuning for closed-loop AD planning. FlashPlanner introduces a continuous future goal point as a compact navigation interface for the generative planning policy. This goal is produced by a continuous goal predictor, which is first pretrained and subsequently optimized with RL using closed-loop feedback from multiple candidate goals at each decision step. To support efficient online reinforcement fine-tuning and real-time deployment, FlashPlanner adopts a data-prediction flow-matching objective and removes redundant architectural components in existing diffusion-based planners. Experiments on the closed-loop nuPlan and interPlan benchmarks demonstrate that \textit{FlashPlanner} achieves state-of-the-art planning performance while delivering 6× faster inference (166 FPS) than the previous SOTA baseline (28 FPS). We will open-source our project.

Low-Rank Adaptation (LoRA) is a popular parameter-efficient fine-tuning (PEFT) method that adapts the large model with few trainable parameters. Federated LoRA extends LoRA to federated learning (FL), enabling clients to collaboratively fine-tune a shared model without sharing their raw data by exchanging LoRA parameters. However, due to distributed data heterogeneity, local LoRA updates are highly inconsistent across clients, hindering robust global generalization. Moreover, common direct aggregation introduces aggregated bias, yielding noisy updates and further harming the global generalization. Existing bias-mitigation approaches often rely on symmetrically alternating LoRA modules, overlooking matrix distinct characteristics during training. Therefore, we propose FLoRA-Chef, an asymmetric Federated LoRA that decides when and how to aggregate LoRA components. FLFire adaptively selects the next aggregation target by tracking cross-clientupdate dispersion. FLSauce applies matrix-distinct reweighting to control diversity and prioritize stability. Experiments on various scenarios demonstrate consistent improvements, yielding an average gain of 6.94% over counterparts, validating the benefit of adaptive asymmetric aggregation for federated LoRA.

Natural sequences often contain signals that move over time, such as translating objects, rotating digits, and drifting storm cells. Selective state space models such as Mamba process such data efficiently, but their recurrent state is updated at fixed spatial coordinates. As a result, evidence from a moving object can be accumulated across inconsistent locations rather than in the object's moving reference frame. To address this problem, we introduce Flow-SSM, a flow-equivariant selective state space model that lifts recurrent memory to candidate motion branches and transports each branch along its corresponding flow before applying the selective update. To handle real-world sequences where broad motion and local deformation often occur at different scales, we further introduce Hierarchical Flow-SSM, which organizes flow-aligned memory in a coarse-to-fine hierarchy rather than a single flat representation. We validate our architecture progressively: demonstrating exact transformation capture on Moving and Rotating MNIST, interpretable flow structures on KTH action videos, and the necessity of our multi-scale design for complex forecasting on the SEVIR weather dataset. Beyond merely improving predictive accuracy, our analyses confirm that the explicit flow alignment and hierarchical communication contribute measurably to the observed gains. These results support transported recurrence as an effective inductive bias for spatiotemporal sequence modeling.

Large language models (LLMs)-based multi-agent systems (MAS) coordinate specialized agents through prompts, tools, and communication topologies, making their hidden workflows valuable intellectual property and security-critical assets. Existing black-box MAS extraction methods implicitly assume that adversarial queries can traverse all agents, which holds for static workflows but breaks down in dynamic workflows whose execution paths depend on input semantics and intermediate states. We identify two key challenges in dynamic workflows: **branch overfitting**, where fully adversarial queries overfit to the same branch and extract only a subset of agents, and **stealthy coverage exploration**, where the adversary needs to achieve complete branch coverage with few redundant queries while not exposing the extraction task. To address these challenges, we propose *FlowLeak* that combines **Task-Preserving Payload Template**, which preserves legitimate task semantics while eliciting workflow information to mitigate branch overfitting, with **Coverage-Guided Branch Exploration**, which uses previously extracted workflow fragments to generate branch-targeted tasks and constrains workflow extraction as an auxiliary task requirement, thereby reducing exploration queries and making extraction harder to identify. Experiments on 102 MAS show that *FlowLeak* addresses both challenges and substantially improves dynamic workflow extraction (2.68$\times$ improvement). Furthermore, we show that the extracted workflow information from *FlowLeak* enhances downstream attacks, highlighting the security risks of MAS workflow extraction and our method.

We reformulate Optimal Transport Conditional Flow Matching (OT-CFM), showing that it admits an exact proximal form via an extended Brenier potential, without assuming that the target distribution has a density. In particular, the mapping to recover the target point is expressed by a proximal operator, which yields an explicit proximal expression of the vector field. We also discuss the convergence of minibatch OT-CFM to the population OT formulation as the sample and batch sizes increase. Using second epi-derivatives of convex potentials, we prove that, for manifold-supported targets, the manifold structure is stable by perturbation to the dynamics: after time rescaling, the dynamics contracts exponentially in directions normal to the manifold while remaining neutral along tangential directions.

Predicting the future motion of multiple people is inherently stochastic: the same observed history admits many plausible continuations depending on intent, coordination, and social context. Yet the dominant paradigm in multi-person 3D motion prediction remains deterministic, with existing methods producing a fixed number of outputs per agent. In contrast, we present a conditional flow matching framework that jointly generates diverse future motions for multiple agents, modeling global trajectories and root-relative poses. A B-spline reparameterization compresses the generative state space while enforcing temporal smoothness, and a dual-stream transformer conditions each agent's predictions on observed neighbor motion for interaction-aware forecasting. Our method achieves state-of-the-art performance on CMU-Mocap (UMPM) and 3DPW, outperforming deterministic baselines on joint position error, pose error, and final displacement error despite producing a full predictive distribution. We additionally establish the first motion prediction baseline on WorldPose, a large-scale professional soccer dataset, where social conditioning demonstrably concentrates predictive uncertainty along interaction-constrained directions.

Functional near-infrared spectroscopy (fNIRS) is a promising modality for brain-computer interfaces and cognitive state decoding, but progress in machine learning for fNIRS has been limited by small, isolated datasets and inconsistent evaluation protocols. We introduce fNIRSAtlas, the largest open benchmark for fNIRS classification to date, comprising 23 public datasets spanning diverse experimental paradigms, including motor execution, motor imagery, mental arithmetic, working memory, and emotion recognition. fNIRSAtlas provides a fully automated pipeline from raw data to evaluation and defines standardized benchmark tasks for both cross-subject and within-subject classification, ensuring reproducible comparison of methods. We evaluate a variety of baseline models from fNIRS and related neuroimaging and time-series machine learning literature on the benchmark. Our results demonstrate that performance varies substantially across datasets and that methods validated on limited data -- especially deep learning approaches -- often fail to generalize beyond their original evaluation. fNIRSAtlas serves as an extensible, open, and reproducible benchmark to support more robust and cumulative progress in machine learning for fNIRS data. Our code and data can be found at https://anonymous.4open.science/r/fNIRSAtlas-8842.

The open-ended generation in LLMs usually requires multi-dimensional rubrics to adequately assess quality and guide the improvement of reinforcement learning. However, a critical dilemma inherent in this training paradigm is the imbalanced reward polarization along different rubric dimensions. Under this bottleneck, even if LLMs achieve relatively high rewards after training, they may still exhibit severe deficiencies in certain dimensions, leading to a direct deterioration in user experience. To address this problem, we propose \emph{Focal Reward}, a novel objective to automatically balance the training of reinforcement learning under rubric-based rewards. Specifically, we first leverage an inverse reward projection mechanism to estimate the saturation degree of each criterion in the rubric, which forms the basis to calibrate the reward direction. Then, the final objective is designed with an automatically reweighting coefficient for each criterion to achieve the fine-grained balancing. Extensive experiments across three model scales and six benchmarks demonstrate that our \emph{Focal Reward} method outperforms the strongest static aggregation baseline in all 18 model--benchmark comparisons. Rollout, mechanism, and ablation analyses further show that these gains arise from online, saturation-aware reallocation toward rubrics that still have room for improvement.

Monocular depth foundation models generalize well across scenes, yet they are typically optimized with uniform pixel-wise objectives that do not distinguish user-specified or task-relevant target regions from the surrounding context. We therefore introduce Focusable Monocular Depth Estimation (FDE), a region-aware depth estimation task in which, given a specified target region, the model is required to prioritize foreground depth accuracy, preserve sharp boundary transitions, and maintain coherent global scene geometry. To prioritize task-critical region modeling, we propose FocusDepth, a prompt-conditioned monocular relative depth estimation framework that guides depth modeling to focus on target regions via box/text prompts. The core Multi-Scale Spatial-Aligned Fusion (MSSA) in FocusDepth spatially aligns multi-scale features from Segment Anything Model 3 to the Depth Anything family and injects them through scale-specific, gated conditional fusion. This enables dense prompt cue injection without disrupting geometric representations, thereby endowing the depth estimation model with focused perception capability. To study FDE, we establish FDE-Bench, a target-centric monocular relative depth benchmark built from image-target-depth triplets across five datasets, containing 252.9K/72.5K train/val triplets and 972 categories spanning real-world and embodied simulation environments. On FDE-Bench, FocusDepth consistently improves over globally fine-tuned DA2/DA3 baselines under both box and text prompts, with the largest gains appearing in target boundary and foreground regions while preserving global scene geometry. Ablations show that MSSA's spatial alignment is the key design factor, as disrupting prompt-geometry correspondence increases AbsRel by up to 13.8%.

Follow the regularized leader (FTRL) is a foundational algorithm in online learning whose regret properties have been extensively studied for decades. However, in stark contrast to mirror descent---and gradient descent in particular---the convergence of FTRL to first-order stationary points in constrained nonconvex optimization was hitherto unresolved. In this paper, we show that continuous-time FTRL can fail to converge even asymptotically in constrained optimization problems; that is, the Karush-Kuhn-Tucker (KKT) gap of FTRL can remain bounded away from zero indefinitely. This is especially surprising in light of the fact that continuous-time FTRL guarantees a monotonic improvement of the objective value. As a result, we establish that non-convergent behavior of no-regret dynamics---which besets general variational inequality problems---can occur even in potential systems. From a technical standpoint, our construction relies on an infinitely differentiable function where the gradient flow dynamics exhibit decelerating oscillations without ever converging pointwise, which we in turn embed into higher-dimensional FTRL dynamics that sustain a perpetually large KKT gap.

Modern forecasting methods are often designed and evaluated on datasets that do not reflect the realities of real-world clinical time-series: inter-patient variability, feature constraints, irregularity, and sparse sampling. Microbiome dynamics modeling represents these challenges, has seen little attention from the machine learning community, and has practical significance in medicine. Moreover, it offers both deep domain knowledge and sufficient data to train deep models, making it a strong testbed for a wide variety of forecasting methods. We propose a unified evaluation protocol for forecasting microbial dynamics and conduct MicroCastBench: a benchmark for models covering a wide range of priors (from mechanistic to foundation models) and assumptions (from per-patient to meta-learning models). We find that meta-learning models lead across most scaling regimes and at longer horizons, with mechanistic priors winning at small cohort sizes, short trajectories, and high feature counts. These results suggest that when the target data is far from pretraining distributions, domain-informed and meta-learning approaches can outperform generalist foundation models, highlighting the value of evaluating on under-represented domains.

Accurate 3D surface reconstruction from unposed images remains a fundamental challenge in computer vision. Recent feed-forward pointmap methods eliminate the need for camera poses but inherently produce discrete, sparse representations. In this paper, we present a generalizable pipeline that directly recovers continuous, high-fidelity 3D surfaces from unposed images by seamlessly integrating pointmap- and ray-based representations. Our approach first extracts explicit geometric priors using a pre-trained pointmap model. A query-view-centric selection module then identifies geometrically relevant reference views via a robust surfel voting mechanism. Next, a cross-attention regressor bridges target query rays with reference features to estimate a continuous coarse surface. Finally, a raylet-based module aggregates local 3D features to carve out high-frequency details and analytical normals. Trained as a single foundation model on 9 diverse datasets, our method demonstrates exceptional zero-shot generalization on 6 unseen datasets, clearly surpassing state-of-the-art baselines in geometric accuracy and detail preservation.

Frame selection---picking a small set of informative frames from a video to feed a vision-language model (VLM)---sits at the front of every video-VLM system, yet the literature is fragmented: most works do not share a candidate pool, encoder, frame budget, or answering VLM, making published gains hard to compare across works. We introduce FrameSelect, to the best of our knowledge the first Python library that treats the selector as a standalone, interchangeable component, cleanly separated from the encoder that computes representations and the VLM that answers. Any registered selector composes with any registered evaluator through a single call, and adding a new selector, dataset, encoder, or VLM backend is an isolated extension; a replay mode further rebuilds any table under a different answering VLM without re-running selection. We release reference implementations of 20 selectors, 15 evaluators spanning answer-level and selection-level metrics, and the first unified benchmark comparing frame selectors under a shared protocol across seven evaluators spanning both evaluation modes. This comparison surfaces findings that existing studies cannot: rankings invert across benchmarks. Our library is available at https://anonymous.4open.science/r/FrameSelect-D2A4.

Gradient-based attribution methods are model-faithful and scalable, but Integrated Gradients (IG) can be brittle because explanations depend on heuristic baselines, straight-line paths, discretization, and saturation. We propose Fisher--Rao Integrated Gradients (FRInGe), which defines both the reference and interpolation schedule in predictive distribution space. FRInGe replaces input baselines with a maximum-entropy predictive reference and follows a Fisher--Rao geodesic on the probability simplex. The corresponding input-space trajectory is realized through the pullback Fisher metric and stabilized by KL and Euclidean trust regions; attributions are obtained by integrating input gradients along this trajectory. Across six ImageNet architectures, FRInGe most clearly improves calibration-oriented attribution metrics, especially MAS scores, while remaining competitive on perturbation AUC and infidelity.

Agents based on large language models are increasingly expected to autonomously handle negotiation and transactions. However, existing benchmarks predominantly focus on text-only, bilateral, zero-sum price haggling, failing to capture the complexity of real-world commerce. To address this gap, we introduce Agenticpay, a unified framework and benchmark for evaluating how well multimodal agents reach high-welfare, multi-dimensional agreements in realistic markets. Built around four core components ( Environments, Tasks, Agents, and Metrics), Agenticpay comprises 160 multimodal tasks spanning 4 real-world business scenarios (E-commerce, Food Delivery, Ride-hailing, and Apartment Rental) and 8 market topologies, scaling from 1-to-1 bargaining to many-to-many (N-to-N) competitive markets. Beyond price haggling, agents must read product images, infer each opponent's hidden preferences, and trade off multiple binding contract terms (e.g., price, lease duration, return policy, delivery speed). Agents communicate through multi-round natural-language dialogue, with each turn proposing or revising a full contract, and outcomes are scored by a utility-based framework that rewards agreements maximizing joint welfare. Evaluations on state-of-the-art proprietary (GPT-5.4, Claude Sonnet 4.6, Gemini 3.1 Pro Preview) and open-weight (Qwen3-VL-32B-Instruct, InternVL3-38B) multimodal models reveal substantial gaps in non-zero-sum value creation and market reasoning: even the strongest agent (Gemini 3.1 Pro Preview) reaches a GlobalScore of only 42.3/100, and performance consistently degrades as markets scale from bilateral to multi-sided, with the average GlobalScore dropping by 5.7 points and open-weight models suffering the largest declines (up to 11.5 points). These findings establish Agenticpay as a foundational testbed for multimodal agentic commerce. Code and dataset are available at: https://anonymous.4open.science/r/AgenticPay-4BBB/README.md.

Neural decoders have shown strong potential for error correction in short- and moderate-length regimes, yet a fundamental tension remains between decoding accuracy and inference latency. Recent attempts to leverage diffusion probabilistic models for channel decoding typically adopt a fully generative paradigm, initializing the reverse process from an observation-agnostic prior, which overlooks a key structural property of channel decoding: the received signal already contains substantial information about the target codeword. In this work, we propose Channel Residual Diffusion Model (ChRes-DM), a principled diffusion-based decoding framework that reinterprets channel decoding as a directed restoration process anchored at the noisy observation. Instead of sampling from a generic prior, ChRes-DM constructs a geometric residual bridge between the received signal and the clean codeword, explicitly modeling the conditional transport induced by the channel and leading to a deterministic Probability Flow Ordinary Differential Equation (PF-ODE) that governs the decoding dynamics. By eliminating redundant stochastic sampling inherent in conventional diffusion models, ChRes-DM enables efficient iterative decoding with flexible step-skipping, offering fine-grained control over the accuracy–latency trade-off. Extensive experiments across representative channel coding benchmarks demonstrate that ChRes-DM achieves competitive or superior decoding performance compared to existing neural decoders while significantly reducing inference iterations, highlighting diffusion-based residual transport as a promising and scalable paradigm for neural channel decoding.

Deep Research Agents (DRAs) answer complex questions by searching the web, checking evidence, and synthesizing conclusions across heterogeneous sources. We introduce a category-theoretic framework for evaluating such agents. The framework treats deep research as a structured mapping from user intent to evidence-grounded conclusions, making retrieval traces, cross-source alignment, and final synthesis explicit. Guided by this view, we build a mechanism-aware benchmark of 296 bilingual questions covering four structural skills: following multi-hop evidence chains, verifying claims across sources, re-ordering fragmented information, and rejecting unsupported assumptions. We evaluate 16 systems with human verification and find that these tasks remain difficult: the best system reaches 19.9\% average accuracy. The results reveal complementary strengths across systems, but also persistent weaknesses in long-horizon retrieval and intersection-heavy verification. We further instantiate two theory-guided interventions, tracked search and category tools, in API-based agents. These variants improve over their corresponding baselines, suggesting that the framework is useful not only for diagnosis but also for modest, targeted system design.

Query-aware video summarization (QVS) aims to select a concise set of moments relevant to a textual query in the context of the entire video. Existing QVS methods are often driven by local query-segment relevance scoring, which is effective for retrieving query-matching moments but insufficient for composing coherent summaries over long videos. Long-form QVS is especially challenging because query-relevant evidence tends to be sparsely distributed across extended timelines, and satisfying a query may require combining complementary segments while avoiding redundancy under a strict summary budget. We propose QLVSumm, a framework that autoregressively generates a query-conditioned video summary, capturing inter-segment dependencies as well as the saliency of the segment and relevance to the query. We also introduce a large-scale benchmark for long-form QVS (LQVS) with open-vocabulary queries and budget-constrained extractive summaries, together with a complementary metric for evaluating query-aware summary disentanglement. Experiments show that QLVSumm achieves state-of-the-art performance and demonstrates stronger query-aware summary disentanglement.

Diffusion language models (DLMs) can generate multiple tokens in parallel, but training large DLMs from scratch remains expensive. A practical alternative is to adapt off-the-shelf autoregressive (AR) checkpoints into diffusion models, reusing their linguistic and reasoning capabilities. Existing adaptation recipes either modify logits and grow attention masks toward full-sequence diffusion, or directly fine-tune ARweights under a block-diffusion objective, leaving two questions underexplored: what diffusion paradigm should AR-to-DLM adaptation target, and what transition path preserves AR knowledge most effectively? We argue that Block-Diffusion is a natural destination because AR decoding corresponds to block size one at the level of attention and generation order, while larger blocks introduce controlled intra-block bidirectionality and parallel generation. Based on this view, we propose a context-causal adaptation path that keeps committed context strictly causal, a one-pass parallel training formulation with auxiliary AR guidance, and a gradual block-size curriculum. Across several AR initializations and model scales, these components improve average adaptation performance over random mask annealing and direct fine-tuning. Scaling the recipe yields NBDIFF-7B, which supports 32K-token contexts and achieves the strongest average performance among the compared diffusion LLM baselines on general, math, and code benchmarks. Code and checkpoints will be released upon publication.

Semi-Supervised Time Series Classification (SS-TSC) with Deep Neural Networks (DNNs) achieves strong accuracy by leveraging unlabeled data, yet the resulting models remain black boxes that offer no insight into which features drive each prediction. Existing post-hoc explanation methods for TSC can identify important features but provide no guarantee that the model genuinely relies on them, while ante-hoc approaches impose architectural constraints that limit flexibility in existing model usages. Neither setting explore the semi-supervised regime, where limited labels can affect explainability. We propose \textbf{Sensory Gating}, the first SS-TSC framework that simultaneously bridges post-hoc attribution and ante-hoc explanation through a four-stage pipeline: (i) a base classifier is trained with our proposed Forecasting Joint-Embedding Predictive Architecture (F-JEPA) auxiliary objective; (ii) post-hoc attribution maps are converted into per-instance binary masks via greedy selection of sufficient features; (iii) a lightweight \textbf{Sensory Gate} amortizes these masks through Binary Cross Entropy (BCE) supervision with data-driven threshold calibration; and (iv) a knowledge distillation stage to provide an explainable SS-TSC student model that operates on gated (masked) input, while consistently reproduces the original black box teacher (F-JEPA)'s predictions. We validate our framework on eight datasets across four label ratios with five seeds per configuration. F-JEPA achieves better performance compared to recent state-of-the-art (SOTA) semi-supervised TSC methods, while the sensory gate reduces the fraction of features the classifier requires to 20--45\% with competitive accuracy. On MIT-BIH, where ground-truth saliency annotations are available, the gate's selected features align with domain-relevant regions while classification performance remains competitive with the ungated baseline at 10--20\% kept ratio.

Training strong large language models (LLMs) requires high-quality supervision, which is often scarce. Recent work shows that paired preference data from weak–weaker model pairs (e.g., Qwen3 4B over 1.7B), despite the limited quality of individual responses, can provide an effective supervision signal through relative quality deltas, which we term a "weak" signal. This motivates a key research question: can multiple "weak" signals be constructively aggregated for improving strong models (e.g., Qwen3 8B)? To this end, we propose Preference Delta Aggregation (PDA), the first framework that derives a preference delta from each weak-weaker model pair, instantiates it as a LoRA adapter learned through preference optimization, and aggregates the resulting deltas via LoRA merging. To further mitigate directional interference during LoRA merging, we introduce Geometric Alignment Merging (GAM), a geometry-aware merging method that aligns adapter subspaces before aggregation, enabling more robust composition of diverse deltas. Evaluations on knowledge reasoning and agentic search benchmarks show that aggregating multiple "weak" signals pushes performance beyond any single signal, with further gains as additional signals are incorporated. Correspondingly, PDA with GAM improves the strong model by 6.8 and 7.3 points on average for knowledge reasoning and agentic search, respectively. It outperforms all single-delta and multi-delta baselines, exceeding the best single-delta baseline by 2.1 and 4.3 points. Further analysis attributes these gains to the effective composition of complementary capabilities encoded across distinct preference deltas.

Ternary Vision Transformers offer substantial model compression, however state-of-the-art methods only ternarize the encoder layers, leaving patch embeddings, LayerNorm parameters, and classifier heads in full precision. In compact models targeting resource-constrained processors, such as microcontrollers, these remaining full-precision components determine the total memory footprint, severely limiting deployment efficiency and on-device feasibility. In this work, we introduce a fully ternarized Vision Transformer in which \emph{all} weight matrices and normalization parameters are ternarized (FTerViT). To this end, we introduce two novel operators : TernaryBitConv2d with per-channel scaling for patch embedding and TernaryLayerNorm. FTerViT is trained using knowledge distillation, followed by a lightweight quantization-aware recovery phase. Our ternary W2A8 DeiT-III-S at 384$\times$384 resolution achieves 82.43\% ImageNet-1K top-1 at 6.09\,MB (${\sim}$15$\times$ compression, $-$2.42\,pp vs.\ FP32), outperforming prior ternary ViTs methods up to 8 pp. Finally, we demonstrate the first implementation of ternary vision transformers on a dual cores XTensa LX7 microcontroller inside the ESP32-S3 system-on-chip. By deploying FTerViT-Small (based on DeiT-III-Small at 224$\times$224 resolution, 5.81\,MB), we achieve 79.64\% ImageNet-1K top-1 accuracy.

We present DuplexMotion, a streaming, full-duplex speech-and-motion model designed for dyadic interaction. To capture the continuous and reciprocal nature of human communication, this full-duplex capability empowers the agent to simultaneously perceive and generate both speech and physical motion in a streaming fashion. At its core, our method leverages the strong priors of a foundational full-duplex speech model and integrates a novel motion pathway, thereby achieving fully synchronized multi-modal interaction. Specifically, we design a dual-tower Transformer architecture that preserves the zero-shot conversational reasoning of a frozen base speech model while constructing a deeply coupled, streaming motion pathway. By introducing a unified dyadic token interleaving mechanism and guiding cross-attention via a time-aligned speech-motion RoPE, our model effectively aligns autoregressive motions with rich latent speech features. Trained on the 4,000-hour Seamless Interaction dataset, our model effectively captures cross-speaker dependencies and establishes new state-of-the-art performance across both monadic and dyadic human interaction benchmarks.

Few-shot CLIP adaptation is often treated as a parameterization problem, where full fine-tuning is assumed to overfit, so parameter-efficient fine-tuning (PEFT) methods are preferred. In this paper, we show that this conclusion conflates parameterization with optimization. Across 11 datasets, 3 CLIP backbones, and 5 shot levels, dense full fine-tuning with plain SGD outperforms representative PEFT methods, whereas Adam/AdamW collapses. We trace this collapse to unreliable support-set preconditioning. The few-shot second moment systematically underestimates the corresponding full-data statistic, already at initialization. Adam’s inverse-square-root normalization converts this error into an oversized adaptive update. Matching Adam’s first-step update budget to SGD resolves most of the collapse, whereas numerator-side alignment does not explain the failure. Shampoo and SOAP also fail, showing that the problem is not merely Adam’s diagonal approximation but a broader mismatch between support-set and full-data preconditioning geometry. Theory formalizes how few-shot estimation error is amplified into unsafe updates. SGD succeeds because it avoids unreliable preconditioners, stays close to pretrained CLIP, and follows broad, connected, low-curvature corridors. These results identify optimizer-induced preconditioning, not dense full fine-tuning itself, as the main source of failure in few-shot CLIP adaptation. Our code is included in the supplementary materials and will be made public.

Learning expressive representations for Boolean logic circuits is a fundamental challenge at the intersection of graph learning and Electronic $\mbox{Design}$ Automation (EDA). Existing Graph Neural $\mbox{Networks}$ (GNNs) primarily rely on topological message passing, which often fails to capture the strict causal dependencies and discrete functional semantics of logic gates. In this paper, we $\mbox{propose} \textbf{FuncFormer}$, a Graph Transformer that incorporates $\textbf{functional simulation}$ not as a proxy task, but as a fundamental inductive bias directly into the representation learning process. Unlike standard GNNs which typically rely on isotropic aggregation, FuncFormer $\textbf{encodes the intrinsic flow of functional propagation}$ by analyzing randomized simulation traces as they evolve through the network. This approach effectively aligns the continuous embedding manifold with the discrete Boolean function space, effectively mitigating structural aliasing. By integrating these deterministic signal trajectories with a scalable dual-path attention mechanism, our model preserves functional consistency across long-range dependencies in both combinational and sequential circuits. Empirical results demonstrate that FuncFormer significantly outperforms state-of-the-art models (e.g., DeepGate4) in Quality-of-Results (QoR) prediction and formal verification tasks, exhibiting robust generalization to unseen circuit scales.

Real-world multi-modal image fusion faces two fundamental challenges: spatial misalignment and complex degradations induced by heterogeneous imaging sensors. These factors are intrinsically coupled, but most existing fusion methods typically address only one of them, inevitably leading to failure under the other challenge. In this paper, we propose G2Fusion, the first fusion framework that simultaneously addresses image registration and information restoration. By developing a novel geometric-to-generative paradigm, it can directly produce high-quality fused images from unregistered and degraded inputs captured by heterogeneous imaging sensors. This framework consists of two key modules: a flow-based geometric deformation reduction module (Flow-GDR) and a DiT-based generative fusion module (DiT-GF). The former reduces large non-rigid discrepancies through dense flow estimation. The latter performs generative fusion, progressively refining residual misalignments while restoring degraded content through iterative denoising. To enable effective interaction among registration, restoration, and fusion, we design two complementary mechanisms. On the one hand, a target distribution mining strategy is introduced to construct a joint objective distribution from registration, restoration, and fusion, effectively guiding the optimization of DiT-GF. On the other hand, we develop a mutual promotion mechanism that establishes a closed-loop interaction between Flow-GDR and DiT-GF by re-estimating the residual deformation between the fused output and the infrared reference. Extensive experiments demonstrate that G2Fusion consistently outperforms state-of-the-art methods in terms of registration, restoration, and fusion.

Graph-level anomaly detection plays a critical role in a wide range of applications, such as fraud detection in financial networks and malicious program detection, yet it remains challenging in real-world scenarios where labeled anomalous graphs are extremely scarce and graph structures are highly diverse. Existing few-shot methods often suffer from severe performance degradation due to the effects of class imbalance and structural imbalance, and they struggle to generalize to structurally deviant or rare anomalies. In this paper, we propose an adaptive few-shot graph-level anomaly detection model with multi-view structure-aware prompting (GADMVP). To mitigate structural imbalance, we construct a graph-of-graphs representation that facilitates cross-graph message passing, capturing both intra-graph semantics and inter-graph dependencies. Building upon these priors, we introduce a structure-aware graph prompting mechanism that extracts and adaptively refines anomaly-relevant prompts from a small set of labeled graphs. Extensive experiments on multiple benchmark datasets demonstrate that the proposed method consistently outperforms state-of-the-art few-shot graph anomaly detection baselines, while exhibiting strong robustness under extremely limited labeled data. The code and implementation details are available at https://anonymous.4open.science/r/GLAD-3456543/.

Self-supervised gait representation learning aims to learn transferable representations that generalize to unseen scenarios. However, existing methods rely solely on visual self-supervision, making it difficult to assess the semantic value of individual unlabeled sequences and limiting their ability to exploit latent semantic cues. To address this limitation, we introduce language priors into self-supervised gait pretraining. We construct GaitLP-1M, a million-scale silhouette-based gait pretraining dataset without identity labels. A two-stage caption generation pipeline further produces gait descriptions from the corresponding RGB sequences. To the best of our knowledge, GaitLP-1M is the first million-scale silhouette gait dataset paired with gait-centric captions. Building on this foundation, we propose GaitLingo, a language-guided self-supervised framework for gait representation learning. Specifically, we design a Caption-Guided Sample Weighting (CSW) mechanism to prioritize informative samples based on sequence-level semantic richness and attribute rarity. We then introduce a lightweight Temporal Adapter (TA) to better capture motion semantics and improve cross-domain robustness. Finally, we propose Soft Relation Distillation (SRD) to transfer caption-derived relational structures into the visual embedding space. Extensive experiments demonstrate that GaitLingo consistently outperforms prior self-supervised gait pretraining methods in the zero-shot setting across six benchmarks. Compared with GaitSSB, it achieves Rank-1 gains of +12.1% on Gait3D and +7.9% on GREW, as well as an average improvement of +9.1% across four CCPG settings.

We present GAMMA, a scalable feed-forward model that reconstructs 4D Gaussian Splatting from monocular videos, enabling dynamic novel view synthesis in real-time. Most existing approaches for dynamic scene modeling require multi-view videos as input or costly per-scene optimization. In contrast, GAMMA learns a unified 4D scene representation from monocular videos and generatively predicts 4D Gaussian primitives within seconds. The key insight of GAMMA lies in its unified Gaussian reconstruction model, which jointly estimates temporally consistent geometry, appearance, and motion from the input video. We further explore its scalability through large-scale training on a comprehensive dataset encompassing both synthetic and real-world videos. The reconstructed GAMMA representation supports interactive scene exploration via real-time rendering across timesteps and local viewpoints. Extensive experiments demonstrate that GAMMA outperforms existing methods in both reconstruction fidelity and efficiency.

Electromagnetic inverse scattering (EIS), which aims to recover object permittivity from measured scattered fields, is central to a wide range of applications, from medical diagnostics to security screening. However, it is a highly nonlinear and ill-posed problem. Despite substantial progress, existing approaches face a fundamental trade-off between physical fidelity, efficiency, and generalization: data-driven methods are fast but tied to training geometries and sensor setups, while physics-based optimization is accurate but computationally expensive. We introduce a Gaussian-Galerkin (GauGal) method that reformulates EIS from a dense point-wise field reconstruction to a compact, primitive-level physical solving. Rather than using Gaussian primitive merely as a material representation, GauGal take it as the computational units for wave scattering. By projecting the continuous EIS scattering equation into a Gaussian primitive space, the material modulation, Green propagation, source excitation, and receiver observation are all realized as primitive-level operators. This yields a compact differentiable forward solver for physics-consistent reconstruction. Our method achieves state-of-the-art accuracy on synthetic and real benchmarks while reducing runtime from over 30 minutes by leading physics-driven baselines to under 10 seconds. It also generalizes significantly better to unseen geometries and sensor configurations than leading data-driven baselines. Overall, this framework establishes an accurate, efficient, and generalizable paradigm for electromagnetic inverse imaging, enabling fast, physics-consistent imaging in practical settings.

Hybrid model-based reinforcement learning (MBRL) integrates lookahead planning with actor-critic optimization for exceptional sample efficiency. However, this decoupled architecture suffers from a critical planner-actor mismatch. Existing policy alignment methods face a structural dilemma: forward KL triggers mode-covering objective conflicts, while reverse KL relies on brittle proxies that bottleneck expressivity. To address this challenge, we propose Gradient-Embedded Alignment Regularization (GEAR), a unified in-sample extraction framework. At its core, we derive a first-order geometric alignment metric from the continuous-time Hamilton-Jacobi-Bellman (HJB) equation to evaluate action evolution against the local value gradient. By embedding this metric as an absolute modulation weight, GEAR eliminates spurious imitation and transforms the inherently diffuse forward KL into a focused, mode-seeking objective. Evaluations on the challenging HumanoidBench suite demonstrate GEAR achieves significantly higher sample efficiency and asymptotic performance than state-of-the-art baselines. The project code is available at https://anonymous.4open.science/r/GEAR-5E80.

Category-level object pose estimation (COPE), capable of generalizing to intra-class unknown objects, has become a core technique for robotic 3D scene understanding. However, existing COPE methods still require labor-intensive recollection of real-world training data for novel object categories, which limits their scalability in practical applications. This paper aims to achieve synthetic-to-real (Syn2Real) generalized COPE, where a model is trained solely on rendered synthetic data and directly generalized to real-world deployments. The central challenge lies in the significant domain gap between synthetic and real-world data, particularly in texture appearance. To address this, we aim to enhance domain generalization by learning domain-invariant representations that capture semantic commonalities among objects within the same category. We introduce 2D and 3D semantic consistency constraints to reduce the sensitivity of feature encoders to domain-specific features. In addition, we propose an end-to-end pose regression framework that performs 2D-3D cross consistency learning, leveraging dense cross-modality fusion to further refine pose estimation. Since simplicity and effectiveness are essential for real-world robotic deployment, our model operates exclusively on global features, yielding a highly lightweight and efficient architecture. Extensive experiments on the REAL275 and Wild6D benchmarks, as well as real-world robotic manipulation scenes, show superior Syn2Real generalization performance of our paradigm.

Bayesian Reinforcement Learning (BRL), a subclass of Meta-Reinforcement Learning (Meta-RL), provides a principled framework for generalisation by explicitly incorporating Bayesian task parameters into transition and reward models. However, classical BRL methods assume known forms of transition and reward models. While recent deep BRL methods incorporate model learning to address this, applying neural networks directly to joint data and task parameters necessitates variational inference. This often yields indistinct task representations, compromising the resulting BRL policies. To overcome these limitations, we introduce Generalised Linear Models in Deep Bayesian RL with Learnable Basis Functions (GLiBRL). Our approach features fully tractable Bayesian inference over task parameters and model noise, alongside exact marginal likelihood evaluation for learning transition and reward models. The permutation-invariant nature of exact Bayesian inference in GLiBRL enables seamless integration with both on-policy and off-policy RL algorithms. We further show that GLiBRL admits a closed-form relationship between the $\mathcal{L}_2$ distance of its task representations and empirical kernel-based correspondence between task samples, which is to our knowledge the first such structural result for online deep BRL. GLiBRL is compared against representative and recent Meta-RL methods, and improves state-of-the-art performance on both MuJoCo and MetaWorld benchmarks by up to 1.8$\times$.

Modeling an opponent’s intent is critical for effective decision-making in non-cooperative, competitive, and general-sum multi-agent reinforcement learning. Existing opponent modeling methods encode intent using an embedding derived from episode information chosen a priori, such as the opponent’s next action or a future environment state, and use this to guide the ego-agent’s behavior. These approaches assume that the chosen information is universally representative of intent; however, we show empirically that this is not the case as intentions are often task- and environment-dependent. To address this, we introduce a task-adaptive opponent modeling framework that learns a performance-driven mixture of multiple intent representations. We further introduce a new intention representation that maximizes mutual information with the ego-agent’s future returns, thereby capturing opponent information that is most directly relevant to performance. Our approach consistently matches or exceeds the performance of state-of-the-art baselines across diverse tasks and yields insights into when and why different opponent modeling strategies succeed.

Molecules in equilibrium follow a Boltzmann distribution, making the underlying energy landscape a physically grounded modeling objective. However, such landscapes are difficult to learn from data and, once learned, hard to sample from. Diffusion and flow-matching models sidestep these difficulties by learning a time-conditional score or transport field between noise and data, losing the energy inductive bias in exchange for a more tractable training objective. We introduce EBMol, an energy-based model (EBM) that restores this inductive bias by learning an atom-additive scalar potential without explicit simulation during training. Our method employs a flow-inspired Restoring Field Matching objective to approximate the energy landscape. We adopt the Mirror-Langevin algorithm for sampling, enabling unified updates of atomic positions and types, and incorporate parallel tempering for inference-time compute scaling. EBMol is the first EBM for 3D molecular generation to achieve state-of-the-art performance on QM9 and GEOM-Drugs. Moreover, we show that the learned energy landscape serves as a principled quality metric for ranking and filtering configurations, and demonstrate controllable generation without retraining through shape-steered sampling via potential composition and zero-shot linker design.

Generative novel view synthesis from sparse input images is rarely all reconstruction or all generation: pixels visible in some source view have a unique correct value modulated only by view-dependent shading, while pixels in disocclusions or beyond the captured volume admit a distribution of plausible completions. Existing generative novel-view-synthesis methods conflate these regimes under a single uniform loss, blurring the line between geometric fidelity and creative hallucinations even when scene geometry is injected through warped point clouds or projected depth. We introduce GenRec, a multi-view flow matching model that builds the reconstruction--generation split directly into its architecture, supervision, and gradient flow. Guided by an observation mask derived from the source cameras and a monocular depth estimator, a flow matching backbone jointly denoises RGB and scene-coordinate maps across all target views, while a pixel-space refinement stage restores high-frequency detail on observed pixels; the same mask gates supervision so regression signals do not contaminate the generative prior. Across RealEstate10K, DL3DV-10K, and Mip-NeRF~360, in both single-view extrapolation and two-view interpolation, GenRec attains the best reconstruction fidelity in observed regions while also surpassing purely generative baselines on perceptual quality in unobserved ones, showing the effectiveness of our approach. The source code and trained models will be made public.

State space models (SSMs) provide an efficient route to global image restoration because they model long sequences with linear complexity. Recent restoration architectures further enlarge the effective context by reordering image tokens so that semantically similar pixels become neighbors in the 1D scan. This strategy improves texture aggregation, but it also exposes a mismatch between the geometry of images and the discretization used by SSMs: adjacent tokens in the reordered sequence can be spatially distant, while the recurrent update still treats them as consecutive samples of a continuous process. In addition, existing prompt-based modulation relies on a finite set of discrete prompts, which is poorly matched to the continuous variation of natural textures and offers no explicit compensation for the low-pass tendency of recurrent integration. We propose GeoMamba, a geometry-aware and continuously modulated SSM for image restoration. GeoMamba introduces geometry-adaptive discretization (GAD), which conditions the SSM time-step on the Euclidean distance between consecutive reordered tokens and attenuates history propagation across spatial jumps. It also introduces a continuous dynamic matrix (CDM), which regresses pixel-wise modulation parameters from feature and gradient cues to reduce prompt quantization and preserve high-frequency details. Experiments on lightweight and classical image super-resolution demonstrate consistent improvements over strong Transformer- and Mamba-based baselines, while achieving better or competitive performance in additional denoising and JPEG artifact reduction settings.

Existing Retrieval-Augmented Generation (RAG) systems struggle with multi-hop reasoning due to a fundamental inability to balance relevance and novelty. Traditional dense retrievers are constrained by surface-level matching, falling into a homogenized "Similarity Trap". Conversely, emerging Graph RAGs attempt to explore novel documents via heuristic entity extraction, but they introduce prohibitive Large Language Model (LLM) overhead and easily drift into off-topic "Novelty Traps". To fundamentally resolve this contradiction, we propose Geometric Gain Graph (G$^3$RAG), the novel zero-token graph construction paradigm for RAG. G$^3$RAG completely discards the expensive heuristic extraction by LLMs, aiming instead to directly quantify the information gain between nodes through the native document feature space. To this end, we designed a concise edge weight criterion, $\cos\theta \cdot \sin\theta$, which naturally maps the trade-off between relevance and novelty into a geometric measurement of directional consistency and orthogonality in the vector space, thereby driving the token-free topological construction of the graph. Furthermore, we introduce a topological penalty to suppress the excessive connectivity of high-frequency hubs, forcing the transient diffusion process toward long-tail peripheral nodes that carry critical indirect evidence. Extensive experiments demonstrate that, while entirely eliminating graph construction costs, G$^3$RAG significantly outperforms state-of-the-art Graph RAG baselines on complex multi-hop QA benchmarks.

Mechanistic interpretability aims to explain a model’s behavior by identifying causally responsible internal structures. Dictionary-based explainers such as sparse autoencoders and transcoders are a primary tool, but their faithfulness under out-of-distribution (OOD) shift has received little systematic attention. We show that distribution shift rotates the subspace that the model actively uses, misaligning the explainer’s dictionary trained on in-distribution (ID) activations. We formalize this misalignment as the faithfulness gap, a geometric distance between the ID dictionary and the OOD-active subspace, and show that it controls OOD faithfulness degradation. To reduce this gap, we propose the Geometry-Adaptive Explainer (GAE), which realigns the explainer's dictionary with the OOD-active subspace while preserving the original feature structure. This requires only unlabeled OOD activations and no gradient updates. We prove that GAE improves over the unadapted ID explainer, with excess loss bounded quadratically by the second-moment shift. Empirically, GAE even matches or surpasses all training-based baselines in causal faithfulness across multiple models and OOD settings.

Multimodal Large Language Models (MLLMs) have shown increasing potential for 3D scene understanding and spatial reasoning from videos. However, even with strong visual geometry priors, the hidden semantics of existing 3D-enhanced MLLMs can drift across views, leading to degraded grounding, captioning, and spatial reasoning performance. The underlying reason can be their geometry injection as additional tokens or additive features, which enriches visual content but fails to define how view-dependent semantic states should transform. To address this issue, we propose Lie Routing Vision-Language Model (LieVLM), a geometry-conditioned Lie routing framework that treats geometry as an operator over hidden semantics. LieVLM first elicits local Lie states from 3D geometry tokens by predicting latent rotation states and routing coefficients. It then converts these rotation states into triplet-wise Lie group actions to route projected 2D visual semantics. Finally, a geometry-aware routed fusion module combines the original semantic path, the direct 3D context path, and the Lie-routed transformation path for language generation. This design preserves pretrained visual-language semantics while allowing geometry to actively correct view-induced semantic drift. Extensive experiments on 3D scene understanding and spatial reasoning benchmarks demonstrate that LieVLM improves view robustness, especially in severe-pose-change regimes, and consistently outperforms additive geometry fusion methods. Our results suggest a shift in 3D MLLM design, i.e., geometry should not merely be injected as context, but should operate on semantic representations.

Knowledge distillation provides shared semantic supervision for local training in federated learning, reducing the tendency of client models to overfit their own data distributions. Existing methods construct a global knowledge repository from client-uploaded local information to establish consensus across clients. However, global knowledge that is fundamentally derived from client data inevitably contains local biases under heterogeneous settings. To address this issue, this study proposes a geometry-regularized cross-silo consensus enhancement method, termed RISE, which improves collapse resistance in federated learning by preserving pre-trained geometric consensus priors during local adaptation. Specifically, RISE introduces two complementary geometry regularizers. The Biased Low-dimensional Subspace Correction module preserves the spectral energy profile of pre-trained representations by constraining the energy distribution of adapted features within the principal representation subspace. Instead of enforcing direction-wise alignment, it discourages excessive energy concentration along a few dominant components, thereby mitigating biased low-dimensional feature collapse. The Manifold Relation Graph Alignment module regularizes the relational manifold geometry by constructing a transferable inter-class topology from pre-trained global class prototypes and aligning client-side class structures with this global relation, thereby mitigating class representation drift and stabilizing global aggregation. Extensive experiments on eight benchmark datasets across two tasks demonstrate that RISE improves cross-client consensus consistency and enhances the generalization ability of the global model, outperforming eight state-of-the-art methods.

Panoramic 3D reconstruction enables the holistic recovery of complete 360° scene geometry, offering significant benefits for robotics and AR/VR applications; however, direct estimation is persistently challenged by inherent spherical distortions. While decomposing panoramas into perspective views effectively circumvents these distortions and leverages established geometric priors, this strategy often fails to preserve global spatial consistency across the decomposed views. To address this fundamental limitation, we introduce an enhanced feed-forward 3D reconstruction framework tailored for the view decomposition paradigm, driven by two key innovations. First, we propose a geometry-aware attention aggregator, featuring a streamlined attention architecture equipped with a novel Geo-Attention mechanism, which explicitly exploits the intrinsic global information to guide multi-view interactions, thereby suppressing geometrically inconsistent matches. Furthermore, to jointly enforce fine-grained local details and global structural coherence, we formulate a hybrid supervision strategy. Specifically, we complement the standard local point cloud loss in perspective space with our newly proposed spatially uniform global point cloud loss in panoramic space, explicitly enforcing holistic constraints within a unified coordinate system. Experiments demonstrate that our method consistently outperforms prior single-view panoramic approaches in reconstruction accuracy, geometric consistency, and visual fidelity.

Large visual world models have become increasingly capable at synthesizing plausible views and camera trajectories, largely because they inherit strong priors from image and video diffusion. Yet in robotics, spatial computing, AR anchoring, and digital twins, a world is useful only if it can be measured and acted upon: appearance plausibility cannot substitute for geometrically coherent structure. This motivates a geometry-native alternative, where the generative state is the feature space of a geometry foundation model rather than RGB pixels or an appearance latent. Existing geometry-space world models validate this direction, but they still occupy a narrow operating range. When target views lie within observed support, full multi-view interaction limits trajectory length; when observations become sparse, geometry-only training lacks the broad visual priors needed to infer structure beyond observed support. We present GeoWorld, a geometry-latent world model that expands this range without returning generation to RGB space. GeoWorld builds a compact geometry latent, conditions it with camera ray geometry, and uses camera-aware tiered attention with relative pose bias to make long-view denoising practical. For sparse-observation extrapolation, a Bridge module translates hidden states from a frozen camera-conditioned video diffusion model into geometry-latent conditioning, borrowing video priors without making the video model the scene generator. Experiments show that GeoWorld scales geometry-space diffusion to long camera trajectories while improving geometric consistency in prior-assisted extrapolation.

Feed-forward 4D Gaussian Splatting (4DGS) offers an efficient paradigm for reconstructing dynamic scenes from monocular videos, bypassing the need for lengthy per-scene optimization. However, existing methods suffer from three fundamental bottlenecks: redundant frame-wise grid-aligned Gaussian representations, suboptimal single-step 2D-to-4D regression architectures, and restrictive dependencies on 3D/4D signals during training or inference. In response to these limitations, we introduce a novel Gaussian-Grounded Query Refinement (GGQR) framework. To reduce representation redundancy, GGQR learns a compact set of Gaussian queries equipped with a newly designed plateau-shaped temporal kernel, enabling single primitives to persistently model motion-consistent regions across extended frames. To alleviate regression ambiguity, these queries are iteratively refined through a Gaussian-grounded attention mechanism operating in 3D space. Specifically, GGQR leverages foundation models to lift 2D observations into a shared 3D space, explicitly grounds the 3D attention in intermediate Gaussian states, and residually updates those Gaussian states. Trained entirely self-supervised on casual videos, GGQR achieves a new state-of-the-art on dynamic reconstruction benchmarks among feed-forward 4DGS methods, delivering superior performance in novel view-time rendering alongside downstream motion modeling tasks.

Large language model (LLM) agents have demonstrated strong capabilities in long-horizon tasks by interleaving reasoning with tool use. However, as these agents scale to complex workflows such as software engineering and open-ended research, context management becomes a fundamental bottleneck: interaction histories grow unbounded, become costly to maintain, and are difficult to reuse across sessions and agents. We introduce \textbf{Git-Context-Controller (GCC)}, a structured context management framework inspired by software version control systems. GCC elevates agent context from a transient token stream to a persistent, navigable memory workspace with explicit operations---\texttt{COMMIT}, \texttt{BRANCH}, \texttt{MERGE}, and \texttt{CONTEXT}, that enable milestone-based checkpointing, isolated exploration of alternative reasoning paths, and hierarchical retrieval of historical context. By organizing agent memory as a versioned file system, GCC allows agents to manage long-term goals, recover and transfer reasoning across sessions, and coordinate multi-trajectory problem solving in a principled manner. Empirically, agents equipped with GCC achieve state-of-the-art performance on both SWE-Bench and BrowseComp benchmarks. On SWE-Bench Verified, GCC improves task resolution by over 13\% relative to strong long-context baselines and outperforms 26 existing open and commercial systems, reaching over 80\% success rate. The project will be open-sourced for the research community.

Predicting tandem mass spectra (MS/MS) from molecular structures represents a central task in analytical chemistry with direct relevance to clinical metabolomics, systems biology, and adjacent disciplines. In this work, we revisit the problem through the lens of object detection on molecular graphs. Molecular fragmentation, a central step in MS/MS prediction, can be approximated as detecting a set of subgraphs (i.e., fragments) and their associated spectral contributions. Existing fragment-based models follow a two-stage paradigm—first generating candidate fragments and then scoring them—analogous to two-stage R-CNNs in computer vision. Towards higher accuracy and faster inference, we introduce GLACIER, a single-stage transformer-based fragment detection neural network for molecular graphs. This unified formulation eliminates the need for candidate enumeration, enabling scalable and globally consistent modeling of molecular fragmentation. GLACIER is faster and more accurate than existing state-of-the-art methods by a significant margin, achieving 62.4% and 62.1% Top-1 retrieval accuracy with and without contrastive finetuning on the MassSpecGym dataset (from 55.3%), and 55.2% and 38.2%, respectively, on the NIST'20 dataset (from 33.5%). Furthermore, GLACIER provides nearly 3-fold inference speedup over existing two-stage models.

Many engineering and scientific fields have recently become interested in modeling terms in partial differential equations (PDEs) with neural networks, which requires solving the inverse problem of learning neural network terms from observed data in order to approximate missing or unresolved physics in the PDE model. The resulting neural-network PDE model, being a function of the neural network parameters, can be calibrated to the available ground truth data by optimizing over the PDE using gradient descent, where the gradient is evaluated in a computationally efficient manner by solving an adjoint PDE. These neural PDE models have emerged as an important research area in scientific machine learning. In this paper, we study the convergence of the adjoint gradient descent optimization method for training neural PDE models in the limit where both the number of hidden units and the training time tend to infinity. Specifically, for a general class of nonlinear parabolic PDEs with a neural network embedded in the source term, we prove convergence of the trained neural-network PDE solution to the target data (i.e., a global minimizer). The global convergence proof poses a unique mathematical challenge that is not encountered in finite-dimensional neural network convergence analyses due to (i) the neural network training dynamics involving a non-local neural network kernel operator in the infinite-width hidden layer limit where the kernel lacks a spectral gap for its eigenvalues and (ii) the nonlinearity of the limit PDE system, which leads to a non-convex optimization problem in the neural network function even in the infinite-width hidden layer limit (unlike in typical neural network training cases where the optimization problem becomes convex in the large neuron limit). The theoretical results are illustrated and empirically validated by numerical studies.

Memory consumption of the key-value (KV) cache remains a central bottleneck for long-context LLM inference. We introduce PRKV, a training-free KV eviction method that estimates multi-hop global token importance by applying Personalized PageRank over attention-induced transition graphs. This enables importance propagation beyond recency or instantaneous attention, without modifying the model. We benchmark PRKV on long-context understanding, sparse retrieval, and chain-of-thought reasoning tasks using multiple LLM families. Under moderate to aggressive compression ratios, including up to 90\%, PRKV achieves competitive task performance against established eviction baselines while reducing peak memory usage with low end-to-end overhead. These results demonstrate that multi-hop global influence estimation yields a favorable accuracy--memory trade-off for scalable long-context inference.

Linear-time sequence models such as State Space Models (SSMs) offer an efficient alternative to self-attention and have demonstrated strong language modeling performance at scale. In practice, however, their asymptotic advantage is often lost at common training and prefill lengths---optimized FlashAttention kernels remain faster than existing SSM implementations up to 8k tokens. In this work, we substantially narrow this performance gap with novel algorithmic improvements: First, we introduce a split-sequence algorithm that decouples the compute tile from the sequence-parallel split length, preserving efficient matrix multiplication tiles while substantially reducing boundary-state traffic and latency. Second, we resolve a longstanding *speed--stability tradeoff* in computing the 1-semiseparable (SS) decay mask, the main non-matmul bottleneck shared by all chunk-wise linear-time kernels (Mamba-2, Mamba-3, GDN, KDA). Existing kernels use the fast *diff segsum* (prefix subtraction), which suffers monotonicity violations and catastrophic cancellation---failures observed in real pretraining (e.g., Nemotron-H); the stable alternative (*direct segsum*) materializes the dense mask and is $\sim30\\%$ slower. Our *hierarchy-aware segsum* decomposes the SS mask into 1-SS diagonal blocks and rank-one off-diagonal blocks, computed via warp-level scans and cross-warp aggregation directly in registers, achieving the stability of direct segsum at the speed of diff segsum. We implement these algorithms using CuTe-DSL with warp specialization, asynchronous TMA, WGMMA, and layout-aware accumulator fragments. Across a large range of sequence length, our H100 kernels reach 91\% of measured memcpy bandwidth, improve prefill time by $4.1\times$ over the official Triton Mamba-2 kernel, and **outperform the highly-optimized FlashAttention-4 starting at 1-2k sequence length**---to our knowledge, the first time a linear-time sequence layer is faster than state-of-the-art attention at the 2-4k sequence lengths used in common pretraining and prefill workloads, where prior linear-time kernels only reached parity at $\geq 8$k. Our hierarchy-aware segsum is $1.35$-$1.40\times$ faster than a naive stable direct-segsum implementation.

Deployed Graph Neural Networks are increasingly subject to post-deployment interventions: a sensitive attribute is flagged for removal, or a spurious correlation is discovered after the model is already in production. In such scenarios, retraining is often infeasible due to a variety of reasons including scale of modern graphs, inaccessible training data, or strict privacy mandates. The challenge is further compounded in graph-structured data, where attribute dependencies propagate through message-passing, spreading unwanted correlations across the entire network. We propose a framework for surgical, post-hoc attribute invariance in GNNs via gradient-free model editing. Rather than retraining or fine-tuning, our method formulates the edit as a subspace-constrained, closed-form optimization: we identify structurally influential nodes that exhibit high attribute sensitivity, isolate the low-rank manifold in activation space where sensitive variation is concentrated, and solve directly for a weight update that decouples the model's decision logic from the target attribute. The update requires no access to original training labels, introduces no additional parameters, and is computed in a single pass. Empirical results demonstrate that our method enforces attribute invariance while preserving predictive utility, while offering 3 orders of magnitude speed-up over existing GNN editing approaches.

Learning from Label Proportion (LLP) is a weakly supervised learning paradigm in which only aggregated label proportions over collections of instances (i.e., bags) are provided, rather than individual labels. This allows classification while preserving privacy or reducing annotation costs. Existing LLP methods, however, have been largely restricted to i.i.d. tabular or image data. To the best of our knowledge, no solution currently addresses graphs, where instances are inherently interdependent through network structure. In this paper, we generalize LLP to the graph domain and study the problem of node classification with label proportions, where only distributional supervision is available for node bags, and the goal is to infer labels for all nodes in the graph. We argue that the lack of node-level supervision is the main challenge for LLP on graphs, and that existing methods based on i.i.d. assumptions fail to exploit topological correlations. To overcome this, we propose GLLP~(Graph Learning from Label Proportions), a framework that leverages Optimal Transport (OT) with a homophily-aware cost to generate soft pseudo-labels for individual nodes. These pseudo-labels provide stronger supervision signals for training Graph Neural Networks. We further establish theoretical guarantees showing the alignment of our cost function with the node classification objective. Extensive experiments on six homophilic graph benchmarks and the Covid real-world dataset demonstrate that GLLP consistently outperforms existing LLP baselines and variants.

Modeling interacting dynamical systems requires capturing spatial interactions alongside long-range temporal dependencies. Graph neural networks (GNNs) provide a natural representation but typically rely on autoregressive rollouts and treat spatial and temporal dynamics separately, leading to error accumulation over long horizons. Existing approaches also focus on local interactions and short temporal contexts, limiting their ability to capture multi-hop dependencies and global structure. We introduce the Graph Mamba Operator (GraMO), a latent-space simulator that integrates state-space models with graph-based interaction learning. In contrast to prior work that sequences nodes or applies spatial and temporal updates in separate stages, GraMO couples graph-based interactions and temporal state updates within a single recurrence. The update is linear in the latent state, with input-dependent coefficients that adapt across regimes. We evaluate GraMO on N-body systems, motion capture, and robotics datasets, achieving the lowest error across benchmarks and the largest gains in long-horizon prediction.

Sparse autoencoders (SAEs) are increasingly used to extract activation directions for inference-time steering, but their standard sparsity objective treats latent features as independent. This prior can be poorly matched to high-level safety behaviors, where refusal and harmful compliance appear to depend on distributed structure in activation space. We introduce Graph-Regularized Sparse Autoencoders (GSAE), a dictionary-learning method that learns safety-steering directions by smoothing SAE decoder vectors over a neuron co-activation graph and applying the resulting direction bank through a two-gate runtime controller. Empirically, GSAE improves selective refusal across JailbreakBench, HarmBench, and XSTest, increasing harmful-request refusal while keeping benign-prompt refusals low. On Llama-3-8B, adding graph regularization to an otherwise identical SAE bank-and-gating pipeline improves $\Delta_s$ by $20.1$ points on JailbreakBench and $16.8$ points on HarmBench. GSAE outperforms activation-steering baselines and black-box guardrails, preserves benign-task performance, generalizes across Llama-3, Mistral, Qwen 2.5, and Phi-4, and remains strong under black-box and gray-box jailbreak attacks.

Inference-time activation steering provides a lightweight mechanism for aligning language model behavior without retraining, yet existing additive interventions often distort activation geometry, alter representation norms, and induce effective-rank collapse, thereby degrading open-ended generation. Recent norm-preserving rotation-based methods mitigate this pathology, but their reliance on one- or two-dimensional concept axes is insufficient for behaviors whose representations are intrinsically multi-faceted and low-rank. We propose Grassmannian Geodesic Steering (GGS), a rank-preserving framework that lifts activation control from vector directions to $k$-dimensional subspaces on the Grassmann manifold. We develop a theoretical account showing that additive steering provably collapses effective rank under strong intervention, that low-dimensional rotations cannot capture distributed behavioral concepts, and that orthogonal transformations along Grassmannian geodesics exactly preserve the singular-value spectrum. Guided by these results, GGS estimates concept subspaces from contrastive difference covariance, steers hidden states via a closed-form Grassmannian geodesic realized as an orthogonal operator, and adaptively modulates intervention strength using a matrix Bingham gate over subspace-valued activations. The resulting method introduces less than $1.5%$ wall-clock overhead per decoding step. Across LLaMA-3.1-8B and Qwen-2.5-7B on six reasoning benchmarks and an AdvBench safety evaluation, GGS consistently improves alignment accuracy while preserving generation quality and refusal behavior, reducing effective-rank degradation by two orders of magnitude relative to additive baselines, and surpasses the current state of the art among training-free inference-time steering methods.

The ground truths in formal mathematics benchmarks are taken on faith. For theorem proving, released formalizations come without formal proofs that certify their correctness. For autoformalization, faithfulness to the informal source is judged only by the benchmark authors, yet such judgments are inherently subjective and lack collective consensus. We show these assumptions fail at scale: our audit of 367 formal statements in ProofNet finds that 204 (56\%) are unfaithful, and over half of those are mathematically false and cannot be proved. We dissect these errors along three axes (provability, logical strength, root cause) and show that they bias evaluation in predictable, asymmetric directions: false ground truths cap theorem-proving signal and compress model gaps, while equivalence-based autoformalization metrics penalize faithful models more than unfaithful ones. We proceed to fix these ``Ground Falses'' and release the corrected benchmark \textbf{ProofNet-Verified}, produced by a semi-automated pipeline combining Lean provability checks, an LLM faithfulness judge, and human review. Applying the same pipeline to six other benchmarks reveals that unfaithful-GT rates span more than an order of magnitude (4.8\% to 60.0\%), confirming that the failure modes generalize while absolute quality is dominated by curatorial process. Our data and code are available at \url{https://github.com/anonymousauthor567/Ground_False}

Machine Unlearning (MU) is an emerging task to remove specified training data from a trained model while preserving its utility, with Gradient-based Unlearning (GU) standing out as a prevalent category due to high scalability. However, standard MU which solely optimizes for clean accuracy, inadvertently damages adversarial robustness on the retain data. Although a few studies extend MU to adversarially trained models, they rely on strong assumptions (e.g., well-conditioned Hessians, smoothness) and costly Hessian-based updates, limiting scalability and efficiency. In this paper, we propose **GUARD**, a scalable and efficient framework that achieves unlearning while preserving retain-data robustness. Having identified the two causes of robustness degradation, we accordingly design *Robustness-Aware Gradient Projection* and *Importance-Guided Parameter Selection*. Both modules rely on a compact adversarially vulnerable dataset that we construct with or without retain data, to enable scalability under varying data constraints. Validated by theoretical analysis and experiments across 6 settings, GUARD outperforms 6 baselines, boosting robust accuracy by up to 6.3$\times$, closely matching the Retrain oracle in both unlearning and robustness.

Simulation-Based Inference (SBI) serves as a vital framework for parameter inference in scientific fields where simulators involve intractable likelihoods, yet while amortized generative models offer rapid posterior estimation, they are often restricted by the specific priors used during training, thereby limiting their flexibility as prior knowledge evolves. To address this prior dependency, PriorGuide was introduced as an inference-time guidance method, but due to its intractable formulation, it relies on Gaussian approximations of the reverse transition kernel and Gaussian mixture model fitting for the prior ratio, both of which introduce systematic bias. Motivated by these limitations, we propose an unbiased test-time guidance framework that leverages Density Ratio Estimation (DRE) to learn a score guidance term, effectively decoupling the inference process from the prior training. Moreover, our framework remains agnostic to the specific density ratio estimators, making it a general and flexible framework for handling prior changes. Experimental results across multiple tasks demonstrate that our method outperforms PriorGuide, achieving superior performance on metrics such as C2ST and MMD in most tasks while maintaining strong robustness even under limited overlap between the training and target priors. Furthermore, we apply our method to Bayesian updating for parameter inference from planetary light-curve data, where it also demonstrates strong effectiveness and robustness.

Computer-use agents hold the promise of assisting in a wide range of digital economic activities. However, current research has largely focused on short-horizon tasks over a limited set of software. A key reason is that creating environments for complex software requires significant time and human effort, and therefore does not scale. To address this, we introduce Gym-Anything, a framework for converting any software into an interactive computer-use environment. We frame environment creation itself as a multi-agent task: a coding agent writes setup scripts, downloads real-world data, and configures the software, while producing (visual) evidence of correct setup. An independent audit agent then verifies evidence for environment setup against a quality checklist. Using a taxonomy of economically valuable occupations grounded in U.S.\ GDP data, we apply this pipeline to 200 software applications with broad occupational coverage. The result is CUA-World, a collection of over 10K+ long-horizon tasks spanning domains from medical science and astronomy to engineering and enterprise systems, each configured with realistic data along with train and test splits. Distilling successful trajectories from the training split yields strong gains at multiple student scales (2B, 9B, 35B). CUA-World also includes CUA-World-Long, a challenging long-horizon benchmark with tasks often requiring over 500 steps. We also apply the same auditing principle at test time: a separate VLM reviews completed trajectories and provides feedback on what remains, improving Gemini-3-Flash on CUA-World-Long from 11.5\% to 14.0\%. We will release all code, infrastructure, and benchmark data.

Randomized Smoothing provides rigorous robustness guarantees for neural networks without architectural constraints, yet its adoption is limited by extreme computational costs. Standard RS requires tens of thousands of model evaluations per input and forces practitioners to commit to fixed sample sizes a priori. In this work, we present a novel meta-learning framework for anytime-valid certified robustness that adaptively deploys computational resources. By using a lightweight meta-learner to predict image-specific priors for a sequential E-process, we achieve a 20-fold reduction in sample complexity compared to traditional methods while maintaining rigorous statistical guarantees. Beyond raw efficiency, we demonstrate how anytime-validity enables adaptively allocating compute based upon application-specific risk thresholds, a form of resource triage impossible under classic certification frameworks. That this is achievable while also providing similar certification performance demonstrates that our approach provides a pathway for real-time, safety-critical certification deployments.

Evaluating agents on the growing number of agentic benchmarks is challenging because they often require complex environments and agent integrations. We introduce Harbor Adapters, a unified evaluation infrastructure for agentic benchmarks. Our work makes three contributions. First, we develop benchmark adapters that port more than 80 benchmarks to evaluate arbitrary agents, and validate them through rigorous code review and parity experiments. Second, we conduct a large-scale evaluation of frontier models across 54 benchmarks, enabling a broader analysis of agent capabilities and failure modes than was previously possible. Third, we introduce Harbor-Mix, a curated set of 100 difficult, diverse, and high-quality tasks refined from the adapted benchmarks. Harbor-Mix preserves the challenge and breadth of large-scale agentic evaluations while being affordable to run; the strongest model resolves only 15.6\% of its tasks. We release the adapters, evaluation results, in-depth analysis, and Harbor-Mix as open-source artifacts to support more reliable and comprehensive evaluation of language-model agents.

Recent image generation models produce photo-realistic results with rich details, but transferring their powerful priors to video restoration remains difficult due to temporal inconsistency from stochastic detail synthesis. Existing video restoration methods improve temporal coherence but often sacrifice visual fidelity, leaving a gap between image-level quality and video-level stability. We propose a generative video restoration method that bridges this gap by enabling temporally consistent detail synthesis from image generation priors. Our method combines noise energy rebalancing to suppress structure-disturbing low-frequency stochasticity, motion-aligned noise warping to propagate details along object motion, a self-supervised denoising encoder to mitigate encoder-induced temporal uncertainty, and decoupled temporal attention to model cross-frame dependencies. Together, these components preserve fidelity, enhance fine-grained realism, and maintain temporal consistency. Extensive experiments show that our method substantially outperforms existing video restoration approaches in visual quality, detail richness, and temporal stability.

Large language models (LLMs) have achieved strong performance on code completion in general-purpose programming languages, but existing repository-level benchmarks focus almost exclusively on software code and largely overlook hardware description languages. We present HDL-RepoBench, consisting of HDL-RepoBench-Train and HDL-RepoBench-Eval, the first benchmark for multilingual hardware code completion at the repository level with functional evaluation. HDL-RepoBench covers three major hardware design coding styles and annotates each completion target with code-structure-level and hardware-oriented semantic labels derived from concrete syntax tree analysis. Post-training billion-scale code LLMs on HDL-RepoBench-Train enables smaller open-weight models to outperform much larger general-purpose models under matched context windows. Beyond aggregate accuracy, our analysis surfaces four findings that distinguish hardware from software code completion: (i) performance varies sharply across HDLs, with HLS consistently easiest and VHDL hardest across functional pass rate and EM/ES; (ii) hardware accuracy is non-monotonic in code-structure depth, in contrast with the monotonic depth–accuracy curve known for software; (iii) accuracy plateaus once the input context exceeds 2,048 tokens, while software completion continues to benefit from longer context; and (iv) certain repository-level retrievers tuned on software underperform a no-retrieval baseline on HDL repositories because their structural and dependency assumptions are violated by hardware code.

LLM evaluations drive which models get deployed, what safety standards get adopted, which research conclusions get published, and how projections of AI's labor-market impact get made. Yet standard confidence intervals ignore variability from judge model choice, model temperature, and prompt phrasing, producing under-coverage that worsens with more data. The omitted variance can shift results enough to reverse conclusions \citep{baumann2025llmhacking, huang2026dropping}; pipelines that fail to average over it leave the surface that ``benchmark hacking'' exploits \citep{singh2025leaderboard}. This paper decomposes LLM pipeline uncertainty into its sources, distinguishes variance that shrinks with more data from sensitivity to researcher design choices, and uses design-study projections to reduce total evaluation error (TEE). Across the demonstrations, naive standard errors are 40 - 60\% smaller than the TEE-corrected SE. Using Chatbot Arena data, we show naive 95\% CI coverage drops as $n$ grows while TEE-corrected coverage holds at 95\%, and TEE-guided pipelines restrict the benchmark gaming surface from 56 to 32 Elo ($K=27$), below the human-leaderboard baseline. We show further that a small pilot recovers honest CIs and projects which design changes most improve precision. Acting on those projections halves MMLU estimation error against the answer key at equivalent cost, and raises per-match agreement with human votes by 7.9 percentage points on Chatbot Arena.

Reinforcement learning with verifiable rewards (RLVR) has become a powerful paradigm for improving language models on reasoning-intensive tasks, but its effectiveness is often limited by exploration. For example, models often fail on hard problems, leaving little useful reward signal. External expert traces offer a natural source of guidance, yet they may also expose reward-relevant content along the critical path to the verifier target, such as final answers, intermediate values, executable implementations, or answer-related entities. This content can create an unintended reward hacking channel, allowing the policy to obtain reward by copying the trace rather than learning the underlying reasoning or agentic behavior. Existing guided-RL methods reduce this risk by using partial trajectories, but they mainly control how much expert information is shown heuristically rather than which parts should be hidden. To this end, we propose Semantic Masked Expert Policy Optimization (SMEPO), a fine-grained semantic masking strategy for expert-guided RLVR. Instead of truncating traces coarsely or revealing them unchanged, SMEPO masks reward-relevant semantic spans along the critical path while preserving the expert’s decomposition, plan, and procedural structure. This turns hard problems from reasoning from scratch into a fill-in-the-blank process: the policy can follow the expert’s problem-solving route, but must still reconstruct the missing values, code, or entities by itself. SMEPO is simple to apply and requires no changes to the reward function or RL objective. Across diverse domains, including math, code, and agentic search, SMEPO improves accuracy by up to 3.2 points over GRPO and reduces training time by up to 4.2$\times$.

Due to context-length constraints, most MLLMs cannot process full-length videos and therefore rely on sampling a subset of frames as input. However, existing sampling methods, ranging from uniform sampling to relevance-based selection, are often driven by a single sampling principle and thus struggle to accommodate the heterogeneous evidential requirements of different queries: some demand a holistic understanding of how the video unfolds over time, while others hinge on fine-grained events within a short temporal window. To address this limitation, we propose Hierarchical Adaptive Frame Sampling (HAS), a two-stage frame sampling framework. In the first stage, Backbone Frame Construction, we apply a Determinantal Point Process (DPP) to sample frames that are both query-relevant and non-redundant. These selected frames capture key moments across the video and form the backbone of the sampling set, providing a foundation for subsequent enrichment. In the second stage, Adaptive Contextual Enrichment, we analyze the temporal distribution of the backbone frames to infer the query type and adaptively allocate the remaining frame budget between Local and Global Context. The Local Context enriches the backbone with fine-grained temporal dynamics and short-range causal relations, whereas the Global Context connects temporally isolated evidence and provides a holistic view of the entire video to support global understanding. Through such a hierarchical two-stage process, HAS effectively addresses diverse query requirements. Incorporated into three leading MLLMs, it demonstrates consistently superior performance across the Video-MME, LongVideoBench, and MLVU benchmarks. We have released our code.

We present, to our knowledge, the first end-to-end FP4 RL post-training, in which both the rollout and training policies, including their forward and backward passes, operate at 4-bit precision. A systematic study reveals that the dominant source of degradation in FP4 RL is not training-side quantization error but rollout activation quantization: outliers stretch the dynamic range so far that a large number of activation values underflow to zero under FP4. Counterintuitively, restoring the training policy to higher precision while keeping the rollout in FP4 makes accuracy worse than full FP4 baseline, exposing rollout–training mismatch as the principal failure mode and ruling out standard pretraining-style fixes. We address this with Rollout Residual Quantization (Rollout-ResQ): a single residual correction term constrained to a hardware-friendly sparsity pattern, added only to the FP4 rollout matmul — a lightweight correction that recovers most of the precision lost to outlier-driven underflow without inflating the rollout’s compute footprint. On Qwen2.5-3B and Qwen2.5-Math-7B, Rollout-ResQ paired with the HiFloat4 (HiF4) format — whose three-level hierarchical scaling preserves resolution under FP4’s tight 4-bit budget — closes the accuracy gap to BF16 from 4.9% to 1.1%, bringing fully quantized FP4 RL within striking distance of full precision. Applied to the open-standard MXFP4, the same recipe narrows the gap from 13.6% to 5.3%, revealing that FP4 format choice is a key factor that determines the ceiling on recoverable accuracy. Together, these results establish HiF4 as the enabling format for end-to- end FP4 RL post-training, and Rollout-ResQ as the activation-side mechanism that makes the gap to BF16 closable.

Recently, a series of works have started studying variations of concepts from learning theory for product spaces, which can be collected under the name high-arity learning theory. In this work, we consider a high-arity variant of sample compression schemes and we prove that the existence of a high-arity sample compression scheme of non-trivial quality implies high-arity PAC learnability.

High-dimensional robotic reinforcement learning (RL) is often bottlenecked by inefficient exploration in large, redundant actuator spaces. We introduce Devyn (Developing synergies), a lightweight framework that learns a structured low-dimensional action parameterization directly during policy optimization. Unlike methods that require precomputed synergies or structured exploration in the full action space, Devyn starts from a randomly initialized latent-to-action map and shapes it online through task interaction. The policy acts strictly in a reduced latent space, while a learnable synergy matrix decodes latent actions into actuator commands. To make this evolving decoder stable and useful for policy learning, Devyn combines slow decoder evolution with structural regularization, reducing latent-action semantic drift while encouraging compact coordination patterns. Across diverse continuous-control benchmarks, including high-dimensional humanoid and overactuated musculoskeletal systems, Devyn improves upon standard SAC and achieves competitive or better performance than recent methods designed for high-dimensional robotic control. Ablations show that the gains do not come from a generic latent-action bottleneck alone, but rather from the slow, structured evolution of the synergy matrix. Our results suggest that learning the action parameterization itself can be a simple and effective route to scaling RL for high-dimensional robots.

Low-Rank Adaptation (LoRA) has emerged as a widely used technique for adapting large language models (LLMs) to new domains, due to its modular design and broad availability on platforms such as HuggingFace. This availability has motivated efforts to reuse existing LoRAs for domain generalization. However, existing methods often rely on explicit task labels or additional training, which are impractical for deployment. Moreover, they typically activate a fixed number of entire LoRA, leading to parameter redundancy or insufficiency that degrade performance. In this paper, we propose HiLoRA, a training-free framework that performs adaptive hierarchical routing over LoRA pools. Drawing on structural properties of LoRA, we define rank-one components (ROCs), where each LoRA rank is treated as an independent unit. For a given input, HiLoRA first adaptively selects a subset of LoRAs and determines their ROC allocation based on Gaussian likelihoods at the sequence level. At the token level, it refines routing by activating only the most informative ROCs. We further provide theoretical guarantees that HiLoRA selects the most relevant LoRAs with high probability. Extensive experiments show that HiLoRA achieves substantial improvements in domain generalization, with accuracy gains of up to 70\% over state-of-the-art baselines, while maintaining comparable inference throughput.

Generative models for drug design which directly produce synthetic pathways have gained significant popularity due to their ability to constrain the search space to synthetically accessible molecules. However, existing methods have focused primarily on de novo molecular design, and rarely start the generation process from known binders. In this paper, we present HELiX: a template-based GFlowNet for localized exploration of chemical space. HELiX learns to partially decompose a given synthetic trajectory to an intermediate state, and then perform forward synthesis in a manner that preserves synthetic accessibility, leading to diverse, high-reward analog generation. Moreover, we prioritize sample efficiency by incorporating Bayesian optimization into the training procedure. We diagnose problems inherent to training GFlowNets with Bayesian optimization and introduce a greedy acquisition strategy which effectively balances between exploration and exploitation without the need for reward shaping. Finally, we show that local exploration is inherently robust to noisy oracle evaluations, a common problem in drug development when using unreliable proxies for binding affinity.

Estimating 6-degrees-of-freedom (6DoF) head pose from a single RGB image remains challenging under strong perspective distortion. We observe that the widely used design choice of rectangular face cropping introduces a geometric inconsistency. This leads not only to image--label mismatch under 2D rotation augmentation but also exacerbates perspective ambiguity, making it difficult to distinguish projection distortion from the actual appearance and pose of faces located near the image periphery. To address this issue, we propose HOGWARTS, a geometry-driven framework that replaces rectangular cropping with homography warping induced by a newly defined virtual camera space. By explicitly defining a virtual camera that shares the optical center with the original camera and warping the input image into this space, our method compensates for off-axis distortion in a geometrically consistent manner. Furthermore, to accurately estimate 3D translation in the virtual space, we introduce a pinhole-camera-based translation formulation that explicitly compensates for projection scale variation caused by the virtual camera transformation. Consequently, HOGWARTS enables more geometrically consistent and robust 6DoF pose estimation across varying image locations. To the best of our knowledge, this is the first study to estimate the 6DoF pose of a target within a virtual camera space explicitly designed to mitigate perspective distortion. Experimental results on the ARKitFace and BIWI datasets demonstrate that HOGWARTS consistently outperforms prior methods under severe perspective distortion as well as in cross-domain settings. Code will be made publicly available.

Spatial transcriptomics (ST) provides spatially resolved gene expression measurements, but most 3D ST datasets are acquired as independently processed 2D sections. This sliced acquisition creates a difficult reconstruction problem: measurements are dense within each section but sparse along the axial direction, and inter-section technical variation can be confounded with true biological change. Existing coordinate-based implicit neural representations (INRs) provide a continuous modeling framework, but standard isotropic coordinate encodings do not explicitly reflect this axial-to-lateral imbalance. We present HoloGene, an anisotropic conditional INR framework for estimating continuous 3D gene-expression fields from pre-registered ST sections. HoloGene first compresses high-dimensional expression profiles into a graph-autoencoder latent manifold. It then uses pseudo-label-derived biological signature conditioning to provide coarse intra-section biological context, while depth-dependent FiLM modulation models axial variation. To reduce slice-specific covariance shifts, we introduce a signature-stratified correlation-alignment regularizer over latent features. Using held-out measured sections as cross-sectional references, HoloGene improves reconstruction accuracy and biological preservation metrics compared with general-purpose INRs and adapted spatial-transcriptomics baselines. Ablation analyses show that decoupled depth conditioning improves axial generalization, while correlation alignment provides the strongest reduction in slice-specific representation leakage. These results support anisotropic conditional modeling as a practical strategy for estimating continuous 3D expression fields from discontinuous ST sections.

We identify a topological obstruction to the standard stable-attractor template for learning local Nash equilibria, also known as first-order Nash equilibria (FONE), in constrained non-concave differentiable games. We consider continuous deterministic learning dynamics that leave each FONE point fixed and aim to attract every initial condition to the full FONE set. Our main message is that this familiar convergence template can fail for a topological reason. Targets with holes, such as loops, cannot be stable global attractors for such dynamics on contractible state spaces. We then exhibit this obstruction in a simple non-concave quadratic zero-sum game on a box. Its exact FONE set is a diagonal six-edge loop together with an isolated point, so the loop creates the required topological obstruction. As a result, no continuous pointwise-stationary learning rule covered by our framework can make this exact FONE set a stable global attractor. We show that degree two is the minimal polynomial degree at which this obstruction can arise. We further amplify the construction to arbitrary homological degree. Finally, by studying the topology in sufficiently small perturbed version of the game, we show that an open positive-volume family of quadratic games retain the same homological obstruction. Thus, even in quadratic zero-sum games, this phenomenon is not a pathology of a single construction; instead, the geometry of projected first-order stationarity can create an intrinsic barrier to stable global FONE dynamics.

While Graph Neural Networks (GNNs) have proven highly effective at modeling relational data, pairwise connections cannot fully capture multi-way relationships naturally present in complex real-world systems. In response to this, Topological Deep Learning (TDL) leverages more general combinatorial representations--such as simplicial or cellular complexes--to accommodate higher-order interactions. Existing TDL methods often extend GNNs through Higher-Order Message Passing (HOMP), but face critical scalability challenges due to the steep complexity overhead of propagating messages through combinatorial structures. To overcome this limitation, we propose HOPSE (Higher-Order Positional and Structural Encoder), a framework \emph{free of message passing layers} that uses Hasse graph decompositions to derive efficient and expressive encodings over \emph{arbitrary higher-order domains}. Notably, HOPSE scales linearly with the size of combinatorial representations while preserving the expressive power and permutation equivariance of the HOMP approaches. Experiments on molecular and topological benchmarks show that it matches or surpasses state-of-the-art performance while consistently achieving speedups over HOMP-based models, opening a new path for scalable TDL. The code is available at https://anonymous.4open.science/r/TopoBench-BF89.

Out-of-distribution (OOD) generalization aims to improve a model’s generalization capacity using only source data, thereby ensuring reliable performance on unseen domains. Cutout, a well-established method that enhances model generalization by randomly masking out square regions of training samples, has shown great success in standard supervised settings. Despite its seemingly strong con- nection to OOD generalization, we observe that simply combining Cutout with existing OOD methods yields only marginal benefits or even degrades performance. Motivated by this empirical finding, we seek a way to make OOD generalization benefit from Cutout. Our method is inspired by reinterpreting the state-of-the-art hyperspher- ical prototype learning as a mutual information (MI) framework. Within this framework, we observe that naively applying Cutout to existing OOD methods is equivalent to estimating MI from a single masked subview, which leaves much of Cutout’s potential untapped. Instead, we apply the MI chain rule to split the original objective into two smaller estimation problems: one encourages the Cutout subview to align closely with the label, while the other drives the full image to capture the residual information unavailable in that subview. This divide-and-conquer strategy fully exploits the Cutout-generated subview while remaining computationally light- weight. Empirically, we demonstrate that our method outperforms competitive baselines on a broad range of OOD benchmarks and achieves superior performance.

Under matched conditions, classifier-head fine-tuning (\textsc{cls}) paradigm consistently outperforms the LLM-as-Judge sequence-to-sequence (\textsc{s2s}) on binary hallucination detection. We trace this gap to the representations induced by the two objectives. Static analysis shows that \textsc{cls} learns a task-specific discriminative direction, pushing hidden states into a clear two-cluster geometry with stronger class separation and linearly decodable label information. By contrast, \textsc{s2s} relies on the pretrained language-modeling head to make decisions. This minimizes the pressure to reorganize representations, leaving the hidden states much closer to the pretrained backbone. Our analysis also reveals that the \textsc{cls} signal saturates before the final layer, after which upper layers form a plateau of statistically interchangeable and largely redundant representations. However, endpoint redundancy alone does not show whether these layers are dispensable during training. Further training-dynamics analysis reveals 1) a lower pretrained substrate, 2) a narrow upper-middle \textbf{InfoWindow} where task information emerges, and 3) a later-synchronizing upper plateau that mirrors this signal. This mechanism motivates us to remove layers above the InfoWindow. By truncating roughly 31\% of the network, we obtain early-exit models that match or exceed full-model performance on Qwen3-4B and 8B. The same depth-fractional prescription transfers to Llama-3.2-3B and Mistral-7B-v0.3. We hope this helps guide efficient task-specific LLM design.

Kalman filtering provides a principled inference framework for linear Gaussian hidden Markov models, but it leaves open a complementary question of representation: what is the minimum amount of information about the observation history that must be retained to support accurate filtering? We show that this question leads naturally to an indirect rate-distortion problem, in which an encoder observes the history of noisy measurements while distortion is evaluated with respect to the latent state. Directly solving this problem is challenging, since the optimization ranges over arbitrary stochastic kernels from a time-varying observation history to an output representation, resulting in an infinite-dimensional formulation. We overcome this difficulty by proving that any feasible encoder can be replaced by a stochastic encoder acting only on the Kalman posterior mean, achieving the same mean-square error with no larger mutual information. The reduced problem admits a closed-form solution in the form of a linear Gaussian blurring mechanism determined by the spectral decomposition of the Riccati solution. This yields an explicit expression for the rate distortion function $R_t(\epsilon)$, showing that any representation with distortion at most $\epsilon$ must retain at least $R_t(\epsilon)$ nats of information. We further use this characterization to derive information-theoretic lower bounds on the population risk of sequential learning and on the prediction error of temporal GP. Numerical experiments corroborate the theoretical predictions.

How can we effectively oversee AI systems that surpass human intelligence? One proposed answer is AI Safety via Debate, an approach in which two AI systems argue opposing positions and then a judge, who can be either a human or a weaker model, decides which one is correct. By facilitating supervisors in providing high-quality training signals, debate aims to encourage truthful and safe behavior from AI systems whose capabilities exceed our own. However, human judgment is not neutral: people often rely on biases which advanced AI may learn to exploit. Drawing on insights from collective intelligence and social cognition, this paper investigates whether debate can guide human and LLM supervisors toward the truth despite their prior beliefs. We introduce and evaluate several protocol configurations, including (1) the use of open-source, proprietary models and (2) human participants as judges (N=507), (3) different methods for eliciting prior beliefs, and (4) variations in claims within the same topic to test whether belief updates generalize across related subtopics. We also explore setups inspired by the wisdom of the crowds, such as the number and diversity of judges, as well as interaction mechanisms including deliberation and voting, designed to better align judgments with the truth. Overall, we find that debate increases accuracy relative to prior beliefs for human judges (from 41.3\% to 57.6\%), whereas the consultancy baseline, in which a single expert argues for one answer, does not. For weaker language models, however, the results are mixed. Debate does not provide similar benefits in most settings and can even degrade decision accuracy. Regarding multi-judge protocols, human deliberation yields the highest accuracy for human supervisors (62.2\%), while for LLMs it mainly benefits those starting from incorrect beliefs, without producing significant improvements in aggregate performance. We provide empirical evidence supporting Debate as a viable path toward scalable human oversight, while casting doubt on the assumption that such benefits translate to weaker AI supervisors, raising concerns about the reliability of fully automated training and control pipelines in high-stakes AI safety contexts.

Neural surrogates promise large speedups over classical solvers for physical dynamics but fail silently at sharp dynamical events such as shocks, fronts, and contact. We present hybrid neural world models for physical dynamics: a recipe for training and deploying multi-horizon surrogates in physical state space, where a single network with continuous horizon conditioning is trained with direct supervision against textbook reference solvers to predict any future state at horizon $T$ in one forward pass. Although no part of the training data, loss function, or architecture supervises discontinuity location, the trained surrogate encodes it implicitly, recoverable from its forward passes alone as a per-trajectory error map that concentrates on shocks, fronts, and contacts, and stays small elsewhere. The map is competitive with or better than standard label-free baselines including deep ensembles, learned error heads, gradient-magnitude indicators, and locally-adaptive conformal prediction, while using only a single trained network and requiring no calibration set or governing-equation knowledge. The recipe supports two operating points. Mode 1 runs the surrogate alone for maximum throughput, with same-hardware CPU speedups of $26\times$ to $72\times$ against textbook solvers on the PDE environments. Mode 2 uses the error map to gate a reference-solver fallback, deferring uncertain trajectories and roughly halving the surrogate's residual error at the default operating point. The recipe applies without modification across reaction-diffusion, compressible Euler, and rigid-body collision dynamics.

Holistic visual tokenizers are fundamental to unified multimodal models (UMMs) as they map diverse visual inputs into a unified representation space. In this paper, we present HYDRA-X, the first UMM framework that unifies image and video tokenization within a single Vision Transformer (ViT). Our design is driven by two core challenges: efficiently injecting spatiotemporal reconstruction capability into a native ViT, and embedding image- and video-level semantic awareness into the latent space. To address the first, comprehensive ablations reveal two key findings: (1) frame-level causal temporal attention suffices for visual reconstruction, whereas full spatiotemporal attention degrades it; and (2) hierarchical temporal compression substantially outperforms single-step alternatives. To tackle the second, we propose a lightweight decompressor that upsamples temporally compressed features under joint image-video teacher supervision, thereby enforcing complementary semantic structures within the compact latent space. Building on this holistic tokenizer, we further propose a principled improvement of the editing pipeline: source-target interaction should occur at the latent level inside the tokenizer rather than at the semantic level inside the LLM, substantially improving editing consistency and accelerating convergence. Instantiated at the 7B dense model, HYDRA-X achieves strong performance across image and video understanding and generation tasks, laying a solid foundation for future unified-tokenizer UMM exploration.

Foundational marked temporal point process (MTPP) models, such as the Hawkes process, often use inexpressive model families in order to offer interpretable parameterizations of event data. On the other hand, neural MTPPs models forego this interpretability in favor of absolute predictive performance. In this work, we present a new family MTPP models: the hyper Hawkes process (HHP), which aims to be as flexible and performant as neural MTPPs, while retaining interpretable aspects. To achieve this, the HHP extends the classical Hawkes process to increase its expressivity by first expanding the dimension of the process into a latent space, and then introducing a hypernetwork to allow time- and data-dependent dynamics. These extensions define a highly performant MTPP family, achieving state-of-the-art performance across a range of benchmark tasks and metrics. Furthermore, by retaining the linearity of the recurrence, albeit now piecewise and conditionally linear, the HHP also retains much of the structure of the original Hawkes process, which we exploit to create direct probes into how the model creates predictions. HHP models therefore offer both state-of-the-art predictions, while also providing an opportunity to ``open the box'' and inspect how predictions were generated.

Reinforcement learning sharpens LLM reasoning, but parameter updates are expensive, forgetful, and opaque; training-free libraries swap gradients for a textual store yet inherit trajectory-level admission and flat structure. We close these gaps with HyperSkill, a Training-Free Omnimodal GRPO framework that keeps the backbone frozen and replaces gradient updates with auditable edits to a Hypergraph-Indexed Skill-Library, where skills are advantage-bearing nodes and compositions are hyperedges. Three mechanisms drive its evolution: a counterfactually verified, step-level distillation admits new skills; moving-average updates with hyperedge growth and pruning follow retrieval feedback; and an advantage-weighted retriever with similarity and capacity gates falls back to zero-skill. Across 11 text/VL/omnimodal benchmarks, HyperSkill beats every baseline by 6.9 F1 (15.6%) without touching a parameter. Our software and data are publicly available.

Recent diffusion models have advanced image-conditioned video generation, but their accessibility raises risks of unauthorized manipulation. Image immunization has emerged as a promising defense for Image-to-Video (I2V) protection, yet existing methods often rely on degrading visual quality or temporal coherence. In this work, we revisit I2V immunization as reducing source-content preservation, shifting the focus from output degradation to weakening source-conditioned content propagation. To this end, we propose I2V-DETACH, an image immunization method that prevents unauthorized I2V generation through source grounding detachment. I2V models preserve source content by propagating source-image cues into video latents during denoising. Accordingly, I2V-DETACH disrupts this process in two complementary ways: it suppresses source-video grounding to reduce source cue propagation, and strengthens intra-video interactions to shift denoising toward video-side dynamics with reduced reliance on the source image. To obtain informative optimization signals, we initialize auxiliary video latents with source-oriented states that remain within the source-conditioned generation regime while deviating from source-preserving trajectories. Extensive experiments across diverse diffusion architectures show that I2V-DETACH substantially reduces source-content preservation, improves over prior immunization methods, and exhibits strong black-box transferability.

Reference-conditioned image-to-image (I2I) editing models provide a promising zero-shot approach to personalized image generation by jointly conditioning on a reference image and a text prompt. We show that recent I2I editing models, including FLUX.1 Kontext and FLUX.2, achieve strong subject identity preservation and prompt alignment, but exhibit seed-wise diversity collapse: under the same reference image and prompt, different random seeds often produce highly similar outputs. We further find that this collapse is associated with over-alignment of latent-token trajectories during denoising. Based on this observation, we introduce Selective Hidden Perturbation (SHiP), a training-free inference-time intervention that perturbs only the attention-output representations of latent tokens during early denoising while keeping reference-image and text tokens unchanged. Across diverse subjects and prompts, SHiP improves seed-wise diversity on both FLUX.1 Kontext and FLUX.2 while largely preserving identity and prompt fidelity. On FLUX.1 Kontext, SHiP increases Vendi Score from 1.546 to 2.066 and LPIPS from 0.475 to 0.621, with only minor decreases in GPT-ID and GPT-Text. Code is available at https://anonymous.4open.science/r/i2i-personalization-diversity-0260/.

Neural circuits are densely interconnected systems, and many neural computations are thought to be mediated by the dynamical evolution of activity within these recurrent networks, a hypothesis often referred to as ‘computation through dynamics’. Under this view, understanding computation within real neural populations depends on building models of the dynamical rules governing the temporal evolution of recorded high-dimensional neural activity. Most commonly, such models depend on explicit generative parameterizations of (i) the (nonlinear) intrinsic dynamical flow field and stochasticity, (ii) the (nonlinear) mapping from dynamical state to neuronal activity, and (iii) the variability of individual neural responses given that dynamical state. Misspecification of any of these components can bias estimates of dynamics and latent trajectories. In this paper, we circumvent generation-related issues by introducing the Recognition-Parameterized Gaussian Process dynamics (RP-GPdyn) model. RP-GPdyn models the dynamical flow field using a nonparametric Gaussian Process-defined transition function and models the relationship of dynamics to neural activity implicitly using the Recognition-Parameterized Model (RPM) framework. We show that this generation-free approach uncovers meaningful, behaviorally relevant latent variables and dynamics from both synthetic and experimental datasets.

We present ImmuVis, a family of efficient foundation models for imaging mass cytometry (IMC), a high-throughput multiplex imaging technology that handles molecular marker measurements as image channels and enables large-scale spatial tissue profiling. Unlike natural images, multiplex imaging lacks a fixed channel space, as real-world marker sets vary across studies, violating a core assumption of standard vision backbones. To address this, ImmuVis introduces marker-adaptive hyperconvolutions that generate convolutional kernels from learned marker embeddings, enabling a single model to operate on arbitrary measured marker subsets without retraining. We pretrain ImmuVis on the largest IMC dataset to date, IMC17M (28 cohorts, 24,405 images, 265 markers, over 17M patches), using self-supervised masked reconstruction. ImmuVis outperforms state-of-the-art baselines and ablations in virtual staining and downstream classification tasks at substantially lower compute cost than transformer-based alternatives, and is the sole model that provides calibrated uncertainty via a heteroscedastic likelihood objective. These results position ImmuVis as a practical framework for real-world IMC modeling.

Operator learning has attracted significant interest for its capability to approximate mappings between function spaces. A common assumption in this field is that training and test samples are drawn from the same probability measure. However, this assumption is often violated, leading to probability measure shifts. Under such shifts, standard operator learning methods often lead to degraded performance because the expected training error does not match the expected test error. In this paper, we propose importance-weighted operator learning (IWOL), a framework for learning operators under probability measure shifts in function spaces. We prove that, when the test measure is absolutely continuous with respect to the training measure, the expected test error is equivalent to an importance-weighted expected training error, with weights given by the Radon--Nikodym derivative. This result provides a theoretical basis for correcting measure shifts in operator learning, but importance weighting relies on this absolute-continuity condition and may yield unbounded weights. To overcome these limitations, we introduce relative importance weights defined with respect to a mixture of the training and test measures. The resulting weights remain well-defined even when absolute continuity fails, and they are uniformly bounded. These weights induce a relative importance-weighted training objective for neural operators under probability measure shifts. To estimate these weights, we present binary classifiers that take functions as inputs, defined through functionals modeled by kernel integral operators. This design enables discretization-invariant weight estimation in function space. Experiments on multiple PDE benchmarks and neural operator architectures demonstrate that the proposed method consistently improves performance under probability measure shifts.

Achieving both privacy and robustness against poisoning attacks remains a central challenge in federated learning (FL), since hiding individual updates makes attacks harder to detect and mitigate. Recently, Mai et al. (NeurIPS 2024) proposed RFLPA, a packed-secret-sharing (PSS)-based realization of FLTrust, which weights client updates by their cosine similarity to a trusted server update. In this paper, we show that RFLPA contains design flaws and mathematical errors in its use of PSS, making the protocol theoretically unsound and practically non-functional. We then propose an improved verifiable FL framework that restores correctness while retaining efficiency. Our key insight is a reinterpretation of the algebraic structure of PSS, which leads to a new transposed degree reduction algorithm and a pairwise verification technique for dot products. These primitives enable local matrix-based operations without complex circuit evaluation and substantially reduce communication rounds compared with BGW-style computation. Building on them, we redesign the verifiable aggregation algorithm of RFLPA by introducing perpendicular vectors, reducing cosine similarity verification to a single dot-product check and eliminating costly verifiable secret sharing during enrollment. The resulting framework strictly improves efficiency, fixes the flaws of prior work, and preserves $N/3$-robustness in malicious settings. Our implementation and empirical evaluation confirm the feasibility of the resulting protocol for high-dimensional FL updates.

Confronted by the existing prototype-based methods that often fail to capture complex intra-class variations and inter-class semantic relationships, weakly supervised semantic segmentation (WSSS) faces persistent challenges due to the inherent incompleteness and boundary inaccuracies of Class Activation Maps (CAMs). To tackle these critical limitations, we propose a relationally optimized prototype Memory bank (RO-PMB) framework in this paper, where the integration of Graph Neural Networks (GNNs) with a dynamically optimized global prototype memory bank is pioneered to explicitly model and learn the intricate semantic relationships among all prototypes. Specifically, our Prototype Relationship Diagram Construction (PRDC) module leverages GNNs for contextual message passing over a learned prototype graph, enriching global prototypes with crucial associative knowledge. As a result, a subsequent context-aware refinement process is sustained to inject structured semantic information into local image-specific prototypes, thereby generating CAMs with unparalleled completeness and boundary precisions. Extensive experiments validate that, compared with the existing representative state of the arts in multi-stage WSSS baselines, our proposed RO-PMB achieves competitive and superior performances on a range of challenging benchmarks, such as PASCAL VOC and MS COCO etc. Code will be released.

The standard in LLM-based prediction is to use the final-layer representation as the input to a downstream predictor. However, intermediate layers may encode complementary task-relevant signals. Existing approaches therefore either search for the best layer for each task or apply expensive attention-based mechanisms to learn inter-layer aggregation. In this work, we first show that such complexity is unnecessary: a lightweight Graph Neural Network over a fully connected graph of LLM layers is more efficient and achieves significantly stronger predictive performance than existing approaches. We then introduce the Cayley-Encoder, which further improves both efficiency and predictive performance by replacing the fully connected graph with a Cayley graph over $SL(2,\mathbb{Z}_n)$. These Cayley graphs provide a mathematically grounded topology that is sparse, regular by construction, and has low diameter. This enables effective communication across layers while constraining the aggregation structure and reducing the risk of GNN overfitting. In an evaluation of Cayley-Encoder across 13 tasks and 9 LLMs, Cayley-Encoder consistently outperforms baselines, achieving improvements of up to 40 percentage points in accuracy, while introducing at most 0.1\% additional parameters relative to the LLM size. We further show that Cayley-Encoder is effective in few-shot regimes. Finally, we show that Cayley-Encoder outperforms LoRA fine-tuning while operating on the frozen LLM. We conclude with an explainability analysis showing that multiple layers contribute meaningfully to the final prediction, supporting our hypothesis.

Fine-tuning Large Language Models under stringent memory constraints remains a significant challenge. Recently, Zeroth-Order (ZO) optimizers have been employed to fine-tune quantized models, avoiding backpropagation and further reducing memory usage. However, quantized ZO optimizers often underperform compared to their full-precision counterparts, especially at low bit-widths. We find that existing quantized ZO methods fail to exploit the inherent low-rank structure of gradient updates, thereby limiting their effectiveness. To address this, we propose the Low-rank Quantized Zeroth-Order optimizer (LQZO), which enhances quantized ZO fine-tuning by leveraging the low-rank structure inherent in gradients. Our method introduces three key innovations: (1) A low-rank parameterization of gradient perturbations using structured matrix factorization. (2) A Binary-Aggregated Gaussian perturbation strategy that employs a Rademacher distribution to constrain higher-order moments and tighten the variance bound of gradient estimates, with no additional computational overhead. (3) An Isotropy-Enhanced Estimation technique that reshapes quantization scale groups into more isotropic structures to diversify exploration directions. We provide a theoretical convergence guarantee for LQZO. Extensive experiments demonstrate that LQZO consistently outperforms quantized ZO optimizers, achieving average performance gains of 2.0%--3.3% across quantization precisions. While extreme quantization inevitably leads to information loss, LQZO significantly improves the trainability of INT2 models and narrows the performance gap compared to prior quantized ZO methods. Furthermore, we demonstrate the cross-modality versatility of LQZO by validating its superior performance on Vision Transformers.

In in-context learning (ICL), a frozen pre-trained model solves tasks by conditioning on a prompt of a few input–output examples, without gradient updates. If the task was present in pretraining but the particular prompt sequence was not, the resulting in-distribution generalization is retrieval-based ICL. Learning-based ICL instead reflects out-of-distribution generalization: the model succeeds on prompts generated by a novel task. Empirically, both forms improve with scale. By analogy to benign overfitting in supervised learning, we call this in-context benign overfitting: larger models more faithfully memorize the pretraining tasks (improving retrieval ICL) while also generalizing better to novel tasks (improving learning ICL). We prove that this phenomenon already arises in a minimal in-context linear-regression feature-selection model. In contrast, standard in-context linear-regression models exhibit a retrieval–learning tradeoff, where the emergence of learning-based ICL coincides with degraded retrieval-based performance.

Neural operators have emerged as powerful surrogates for scientific computing, enabling rapid reconstruction of complex physical fields from sparse sensor observations. Despite their strong predictive capability, their reliability often degrades under noisy or corrupted measurements, limiting practical deployment in real-world sensing environments. We show that enforcing statistical independence in latent representations provides a simple yet effective mechanism for improving neural operator robustness. To this end, we introduce Neural Operator Independence Regularization (NOIR), an end-to-end training framework that promotes statistically independent latent features during operator learning. Unlike structured basis approaches such as POD-DeepONet and PCA-Net, which impose orthogonality primarily as an offline compression constraint, NOIR directly shapes latent representations during training to enhance resilience against input perturbations. Across three canonical PDE benchmarks and five additive sensor corruption settings, NOIR consistently improves reconstruction accuracy while preserving clean-data performance. Beyond predictive gains, the induced latent representations exhibit distinctive statistical signatures that reveal architecture-specific robustness characteristics under corruption. These findings identify latent statistical independence as a principled inductive bias for robust neural operator design and provide new insight into the internal structure of operator learning systems. All models, datasets, checkpoints, and training code are publicly available at \url{https://github.com/t55176853-cmyk/NOIR/}

Text- and goal-conditioned human-object interaction (HOI) provides an intuitive interface for specifying high-level intents in controllable character animation and embodied AI. Mapping such intents to physically plausible, contact-rich behaviors naturally requires hierarchical planning and control, yet existing planner-controller systems remain largely open-loop and lack adaptation to physical execution feedback. To address this, we present InfCoiL, a closed-loop framework for physics-based HOI from coordinated planner-controller learning. InfCoiL integrates a morphology-aware flow-matching planner with a unified multi-morphology controller in an autoregressive plan-and-execute loop, enabling diverse, controllable, and executable interactions across diverse objects and morphologies of humanoid embodiments. To further mitigate planner-controller distribution shift, we alternatively fine-tune both components under closed-loop reinforcement learning. Particularly, we introduce Semantic Flow-GRPO, which optimizes the flow-matching HOI planner with semantic covariance over heterogeneous HOI features and group-relative rewards from physical rollouts, while training the controller for reliable execution under contact-rich dynamics. Extensive experiments demonstrate that InfCoiL achieves state-of-the-art performance in both motion quality and task success. The code will be released upon acceptance.

Safety-aligned Large Language Models (LLMs) remain vulnerable to interventions during inference that redirect generation toward harmful outputs. Recent work attributes this to shallow safety, where alignment concentrates in the first few output tokens. We show that shallow safety is a special case of a broader inference-time vulnerability, in which short token injections at any generation step can substantially alter subsequent safety behavior. We also find that a model's alignment with refusal directions in its hidden states does not predict its robustness to such injection, revealing that internal state alone does not determine generation behavior under perturbation. To address this, we align models directly on generation trajectories constructed by simulating mid-sequence perturbation, and show that this improves robustness to mid-sequence injection and generalizes to attacks that exploit early-token generation. Our work argues that robust safety alignment requires training on the generation process itself, not only its outputs.

While the approximation properties of single-layer Transformer architectures have been studied in recent works, a rigorous theoretical understanding of the multi-layer setting remains limited. In this work, we establish that multi-layer Transformers possess fundamentally different approximation capabilities from single-layer ones: for certain retrieval tasks, any single-layer Transformer requires least $\Omega (\varepsilon^{-k})$ parameters to achieve precision $\varepsilon$, where $k$ grows linearly with sequence length $T$, whereas a two-layer Transformer with a single head per layer achieves the same approximation precision with at most $O (\varepsilon^{-1})$ parameters. To understand this separation, we identify two structural mechanisms underlying multi-layer approximation. Specifically, softmax attention can only efficiently retrieve the token attaining the maximum attention score, incurring exponential-in-length parameter cost for $k$-th largest retrieval with $k \geq 2$. Moreover, the parameter cost of decoding coupled information scales with the size of the retrieved token set. Motivated by these findings, we propose **InfoFlow**, a framework for multi-layer Transformers. The framework tracks an information set of accessible input positions at each token and layer, assigning an explicit approximation rate to each mode of information propagation. This abstraction recovers known approximation bounds, remains consistent with experimental observations on trained networks, and yields concrete predictions in settings where direct theoretical analysis is currently intractable. Our results provide a principled framework for reasoning about the approximation efficiency of multi-layer Transformers.

Attribution methods are extensively used to identify which parts of the input are important for a model's decision. However, despite recent efforts there is still no consensus on what exactly they capture and how to evaluate them. We start addressing this problem by explicitly modelling the implicit assumptions that sit at the foundation of attribution methods. We define two types of importance contributions an input region can have. We then create a framework that decouples the process of verifiably establishing contributions from that of evaluating attributions. This shows that current attribution methods struggle to reliably capture input contributions. Further investigation shows a stark difference in attribution performance between model behaviours, exposing a previously overlooked aspect in the attribution literature. Our work calls for clear, quantifiable statements about what attribution methods aim to capture, along with rigorous evaluation frameworks.

Solving algorithmic tasks that involve many sequential steps typically requires scaling up transformer depth or chain-of-thought length, incurring unfavorable parameter and inference-compute costs, respectively, as tasks deepen. Looped transformers promise to recover that depth at a fraction of the cost by reusing a single block across iterations, but it remains unclear how the loop actually performs its computation, or whether it merely simulates a deeper untied stack. We test this on \emph{directed permutation traversal} (\dePo), a synthetic task in the physics-of-LMs lineage in which solving an $H$-hop query requires following $H$ pointers along a fixed permutation of $N{=}16$ entities, with $H$ up to $15$. We demonstrate that, above a minimum effective depth, a $3.2$\,M-parameter looped transformer trained with hop length sampled uniformly per batch traverses all $15$ hops across all seeds, while a $25.8$\,M-parameter fixed-depth baseline at the same effective depth fails to exceed three hops; weight tying, not effective depth, predicts capability. In the same models, linear probes on the residual stream uncover a clean \emph{staircase}: at iteration $k$ the stream linearly decodes hop $k$, with slope exactly one hop per iteration, and each hop stays decodable for roughly two further iterations before being overwritten. The model's own logit lens, applied through the trained final layers, traces the same one-hop-per-iteration ordering at lower amplitude. Taken together, our results recast looped transformers as iterative program learners: weight tying forces the optimizer to discover a single per-iteration update rule, and that rule corresponds to one algorithmic step of the underlying task. Code is available at \url{https://anonymous.4open.science/r/insidetheloop-3ED0}.

AI systems have become increasingly capable of dangerous behaviours in many domains. This raises the question: Do models sometimes choose to violate human instructions in order to perform behaviour that is more useful for certain goals? We introduce a benchmark for measuring model propensity for instrumental convergence (IC) behaviour in terminal-based agents. This is behaviour such as self-preservation that has been hypothesised to play a key role in risks from highly capable AI agents. Our benchmark is realistic and low-stakes which serves to reduce evaluation-awareness and roleplay confounds. The suite contains seven operational tasks, each with an official workflow and a policy-violating shortcut. An eight-variant shared framework varies monitoring, instruction clarity, stakes, permission, instrumental usefulness and blocked honest paths to support inferences regarding the factors driving IC behaviour. We evaluated ten models using deterministic environment-state scorers over 1,680 samples, with trace review employed for audit and adjudication purposes. The final IC rate is 86 out of 1,680 samples (5.1\%). IC behaviour is concentrated rather than uniform: two Gemini models account for 66.3\% of IC cases and three tasks account for 84.9\%. Conditions in which IC behaviour is indispensable for task success result in the greatest increase in the adjusted IC rate (+15.7 percentage points), whereas emphasising that task success is critical or certain framing choices do not produce comparable effects. Our findings indicate that realistic, low-nudge environments elicit IC behaviour rarely but systematically in most tested models. We conclude that it is feasible to robustly measure tendencies for dangerous behaviour in current frontier AI agents.

Large language model (LLM)-driven agents are proposed for a wide range of applications. As scenarios become increasingly composite, a graph-structured multi-agent system (MAS) offers a promising solution due to the orchestration of various skills. To architect a reasonable underlying workflow, recent advances mostly investigate two structures: sequential topology and decentralized topology. The former constrains the structure into a directed acyclic graph for multi-step executions but lacks sufficient scalability, while the latter leverages node-wise subgraphs for knowledge aggregation with synchronous parallelization but lacks long-term planning. Therefore, our key insight is that their strengths are mutually complementary. Driven by this motivation, we introduce step-aware hybrid topology for a MAS, where sequential connections support inter-step ordered planning and decentralized connections promote intra-step perspective integration. Concretely, to effectively coordinate different structural components, we clarify the corresponding topological definition and the protocols for step-wise serialization and parallelization. Subsequently, to dynamically refine desired structures, we introduce the parameterized distribution of a hybrid topology, which is initialized by LLM-driven graph planning. Eventually, to further reduce token consumption and time complexity, we supplement the topological density and length normalization for end-to-end learning. Extensive experiments demonstrate that the hybrid topology outperforms single-form topology in 90.4\% of tasks, our agentic initialization improves learning in 88.8\% of scenarios, proposed normalizations reduce token consumption by 20\%$\sim$30\% and time consumption by 7\%$\sim$13\% with a slight utility fluctuation. The code is available at https://anonymous.4open.science/r/HTAS-D107.

We propose the task of Interactive 4D Volumetric Liquid Forecasting: predicting the spatiotemporal evolution of a future dense 3D liquid velocity field conditioned on how a solid object moves through it. This task provides an upstream physical forecasting primitive for systems that need to reason about motion-conditioned liquid response, including robotic liquid manipulation, digital twins, and rigid-fluid video generation. Existing works have made efforts in predicting flow around static or prescribed solid geometries and synthesizing visually plausible fluid motion. Yet these approaches provide limited coverage for interactive liquid dynamics, as they are not capable of forecasting how a moving solid induces a future full-domain liquid response. In this forecasting scenario, the hydrodynamic effect is condensed near an evolving liquid-solid interface, but the prediction target is a global spatiotemporal rollout. To address this challenge, we propose a two-level recurrent forecasting framework, TIDE/TIDES: TIDE isolates localized interface-driven forecasting, while TIDES adds transport-aware coupling and explicit stabilization to reduce accumulated rollout error. Across adapted neighboring baselines, controlled component ablations, motion/size OOD tests, and long-horizon evaluation, TIDES achieves the best forecasting performance and the ablations identify localized interaction, transport coupling, and stabilization as the key contributors. Together, our task formulation, benchmark, and TIDE/TIDES study establish a controlled foundation for moving-solid-conditioned dense liquid forecasting. Code and benchmark will be made publicly available.

Emerging latent reasoning paradigms allow Large Language Models to “think” in latent embedding spaces, offering a more efficient alternative to explicit Chain-of-Thought. However, they generally face two critical limitations: First, relying exclusively on latent tokens amplifies uncertainty due to their inherently high entropy, thereby degrading solution correctness. Second, the overthinking problem remains prevalent, and existing approaches typically rely on rigid heuristics for stopping, often resulting in suboptimal termination. To address these challenges, we view efficient reasoning as a learnable control problem over both how to think and when to stop. We formulate this within a Reinforcement Learning framework, training lightweight per-step gates on a frozen LLM. This modulation enables the model to (i) adaptively interleave latent soft-token reasoning with explicit token generation, and (ii) predict cumulative exit probabilities for robust early termination. Experiments on diverse reasoning benchmarks across varying model scales and families demonstrate that our method surpasses both standard CoT and existing latent baselines, achieving accuracy gains of ∼3% while reducing token consumption by up to 40%.

Although numerous algorithms have been proposed for unsupervised anomaly detection, there is currently no widely acknowledged internal evaluation method tailored specifically to this task. Existing internal evaluation methods typically assess score distributions alone or depend on auxiliary classifiers, limiting their reliability and interpretability. To address these problems, we propose SSSD (Similar Scores, Similar Data), an evaluation framework inspired by the smoothness assumption in machine learning—that similar inputs should yield similar outputs. We adapt this principle to anomaly detection by requiring that locally similar data points be assigned similar anomaly scores. Instead of testing this assumption directly, SSSD identifies violations of this principle by comparing the similarity of data with similar anomaly scores, interpreting a significant difference as a higher probability of erroneous detection. This evaluation of similarity is measured through nearby-score and nearby-distance, which is calculated based on the distances between neighboring data points and their corresponding anomaly scores. SSSD is highly interpretable, providing valuable insights for identifying mislabeled data. Experimental results show that our proposed method is both effective and robust for evaluating the performance of unsupervised anomaly detection algorithms across artificial and real-world datasets.

Modern image encoders achieve high generalization by decoupling semantic meaning from resolution, an ability yet to be fully realized in the 3D domain. We investigate the failure of 3D point cloud encoders to achieve similar generalization and find that existing models are highly sensitive to sampling resolution and scale changes, leading to significant performance degradation. This sensitivity is a major bottleneck for real-world deployment in robotics, as it suggests models overfit to specific quantization densities and object scales rather than learning invariant semantic features. To mitigate this dependency, we propose Invaria, a point cloud encoder that achieves scale and density invariance through next-resolution prediction and receptive field calibration. While our objective is not the explicit generation of high-resolution point clouds, we find that this training objective encourages the model to learn robust, structural invariants. The resulting encoder achieves significant performance gains during resolution shifts while maintaining high efficiency through a compact model size and reduced token requirements. Specifically, on ScanNet, Invaria achieves a 56.0\% higher mIoU at 3$\times$ lower resolution and a 20\% improvement when the objects scale is reduced by a factor of 3. These gains are achieved with a 45\% smaller model size and an average reduction of 40\% in input tokens. The code will be released once the paper is published.

Self-play fine-tuning enables large language models to improve beyond supervised fine-tuning without additional human annotations by contrasting annotated responses with self-generated ones. Many existing methods rely on a fixed divergence regime. SPIN is closely related to a KL-based regime, SPACE to a Jensen-Shannon-style objective via noise contrastive estimation, and SPIF to $\chi^2$-regularized self-play. Since these divergences exhibit different strengths depending on the distributional gap between model and target, no single choice appears to provide favorable learning dynamics across training stages. We propose IRIS (Interpolative R\'enyi Iterative Self-play), a R\'enyi-based self-play fine-tuning framework with a continuously adjustable objective. IRIS decomposes into two independent tilted risk terms over annotated and synthetic data, with exponential importance weights controlled by the order parameter $\alpha$. We show that several self-play objectives can be interpreted as limiting or representative regimes at particular values of $\alpha$, providing a unified theoretical perspective on these methods. An adaptive order schedule further adjusts $\alpha$ to the distributional gap, shifting from sharper importance weighting early in training to smoother refinement near convergence. Theoretically, we establish the fixed-point property of IRIS and analyze how $\alpha$ controls gradient concentration. Experiments on Zephyr-7B and Qwen2.5-3B across ten benchmarks show that IRIS improves upon baselines, reaching 44.57\% average score with gains across iterations. In our setting, IRIS with only 26$k$ annotated samples surpasses standard supervised fine-tuning trained on the full 200$k$ dataset.

Long-tailed out-of-distribution (LT-OOD) detection is often addressed with specialized training, including auxiliary out-of-distribution (OOD) data, abstention heads, contrastive objectives, energy losses, or gradient-conflict control. We show that these training mechanisms can obscure a simpler issue: frozen long-tailed representations may already contain useful OOD evidence, but raw Mahalanobis distance is distorted by frequency-coupled feature radius and poorly supported tail covariance. We propose \emph{Hyperspherical Pooled Mahalanobis} (\HPM), a post-hoc detector that normalizes features onto the unit sphere and replaces class-specific covariance with a pooled, ridge-regularized metric while keeping class means as semantic anchors. In CIFAR-LT experiments and an ImageNet-100-LT near-OOD boundary analysis, \HPM improves raw Mahalanobis scoring; for Prior-Calibrated ERM (PC-ERM), it raises AUROC from 46.49 to 85.67 on CIFAR-10-LT and from 50.40 to 78.35 on CIFAR-100-LT. This simple PC-ERM+\HPM pipeline also achieves the best Log Efficiency Score (LES; 3.08) on CIFAR-100-LT, retaining roughly 95\% of the best CIFAR-100-LT AUROC observed among the compared post-hoc scores at substantially lower training-time cost. These results argue for evaluating representation quality, detector geometry, and training complexity as separate factors in LT-OOD detection.

Current sign language recognition and translation benchmarks are commonly collected in controlled settings using fixed or studio-style cameras, particularly limiting their ability to reflect how signers communicate in everyday mobile scenarios. As such, this gap becomes increasingly evident when considering the Arabic Sign Language (ArSL), which is a primary sign language used across Arab countries. In this work, we introduce Isharah-Selfie, a large-scale ArSL dataset collected entirely using front-facing smartphone cameras under unconstrained selfie conditions. The dataset comprises 8,997 video clips performed by 16 signers across 1,493 unique sentences, covering several domains such as healthcare, education, and transportation. Unlike conventional datasets dominated by fixed-camera, two-handed signing, Isharah-Selfie captures realistic mobile communication characterized by one-handed signing, close-range viewpoints, partial body visibility, camera motion, unconstrained environments, and heterogeneous device resolutions. Each video is annotated with both gloss sequences for continuous sign language recognition (CSLR) and Arabic sentence translations for sign language translation (SLT). Moreover, we define signer-independent and unseen-sentences evaluation protocols to assess generalization across unseen signers and novel sentence compositions. Extensive benchmarks using RGB-based and pose-based data show that current CSLR and SLT methods remain challenged by selfie-style signing conditions, particularly under unseen-sentence evaluation, where compositional generalization remains weak. By releasing Isharah-Selfie, we aim to support research toward robust, user-centered sign language technologies that operate reliably in realistic smartphone-based communication settings. The Isharah-Selfie dataset is available on https://www.github.com/.

Current Large Language Models (LLMs) are gradually exploited in practically valuable agentic workflows such as Deep Research, E-commerce recommendation, job recruitment. In these applications, LLMs need to select some optimal solutions from massive candidates, which we term as \textit{LLM-as-a-Recommender} paradigm. However, the reliability of using LLM agents for recommendations is underexplored. In this work, we introduce a \textbf{Bias} \textbf{Rec}ommendation \textbf{Bench}mark (\textbf{BiasRecBench}) to highlight the critical vulnerability of such agents to biases in high-value real-world tasks. The benchmark includes three practical domains: paper review, e-commerce, and job recruitment. We construct a \textsc{Bias Synthesis Pipeline with Calibrated Quality Margins} that 1) synthesizes evaluation data by controlling the quality gap between optimal and sub-optimal options to provide a calibrated testbed to elicit the vulnerability to biases; 2) injects contextual biases that are logical and suitable for option contexts. Extensive experiments on both SOTA (Gemini-{2.5,3}-pro, GPT-4o, DeepSeek-R1) and small-scale LLMs reveal that agents frequently succumb to injected biases despite having sufficient reasoning capabilities to identify the ground truth. These findings expose a significant reliability bottleneck in current agentic workflows, calling for specialized alignment strategies for LLM-as-a-Recommender. The complete code and evaluation datasets will be made publicly available upon acceptance.

Image--text contrastive pretraining has become a dominant paradigm for visual representation learning, yet existing methods often yield representations that remain partially organized by modality rather than by semantics. We propose \textbf{ITO}, a framework addressing this limitation through two complementary mechanisms with distinct roles. \emph{Multimodal multiple alignment} enriches supervision by constructing diverse cross-modal correspondences from multi-view image augmentations, providing the primary source of discriminative gain. A lightweight \emph{training-time multimodal fusion} module then acts as a geometric regularizer, encouraging the encoders to produce features that are compatible under fusion and thereby reducing modality-induced separation. Crucially, the fusion module is discarded at inference, preserving the efficiency of standard dual-encoder architectures while incurring training-time overhead only. Extensive experiments across pretraining scales from millions to billions of image--text pairs show that ITO consistently outperforms strong baselines on classification, retrieval, and multimodal benchmarks. Our analysis further reveals that beyond accuracy gains, training-time fusion plays a stabilizing role in optimization, mitigating the late-stage overfitting commonly observed in aggressive contrastive learning.

Large Language Models remain vulnerable to jailbreak attacks despite extensive efforts to ensure safety alignment. Streaming generation scenarios exacerbate this vulnerability by exposing harmful tokens to users immediately upon generation, a phenomenon known as prefix exposure. Existing defenses often fail to address this real-time constraint or impose substantial latency that degrades user experience. To bridge this gap, we propose JEDI, a real-time defense method that secures streaming outputs via in-generation detection and intervention. JEDI leverages representation engineering to monitor the LLM’s internal semantic drift, using a Cumulative Sum algorithm to identify persistent, harmful intent before it manifests in the output. Upon detecting a risk, the system dynamically injects a steering vector to redirect the generation trajectory toward a safe subspace. Extensive evaluations across 6 distinct models and 11 attack vectors demonstrate that JEDI achieves defense success rates ranging from 93.6\% to 99.4\%, outperforming state-of-the-art baselines. Furthermore, JEDI preserves model utility with a Time To First Token overhead of approximately 0.003 seconds, validating its viability for latency-sensitive applications. Our code and experimental results are available at: https://anonymous.4open.science/r/JEDI-93F9

Post-training of Large Language Models (LMs) often prioritizes accuracy and helpfulness at the expense of diversity. This creates a tension: while post-training improves response quality, it also sharpens output distributions and reduces the range of ideas, limiting the usefulness of LMs in creative and exploratory tasks such as brainstorming, storytelling, or problem solving. We address this challenge with Diversity-Aware Reinforcement Learning (DARLING), a framework that jointly optimizes for response quality and semantic diversity. At its core, DARLING introduces a learned partition function to measure diversity beyond surface-level lexical variations. This diversity signal is then combined with a quality reward during online reinforcement learning, encouraging models to generate outputs that are both high-quality and distinct. Experiments across multiple model families and sizes show that DARLING generalizes to two regimes: non-verifiable tasks (instruction following and creative writing) and verifiable tasks (competition math). On five benchmarks in the first setting, DARLING consistently outperforms quality-only RL baselines, producing outputs that are simultaneously of higher quality and novelty. In the second setting, DARLING achieves higher pass@1 (solution quality) and pass@k (solution variety). Most strikingly, explicitly optimizing for diversity catalyzes exploration in online RL, which manifests itself as higher-quality responses.

Tool-augmented language models are bounded by the APIs humans bothered to write; existing tool-creation systems patch this by prompting a frozen LLM at inference time, leaving the model that writes a tool decoupled from the one that uses it, with no signal that the schemas it produces are schemas it can invoke. We propose **SMITH** (Schema-grounded Multi-task Iterative Tool Honing), a reinforcement learning framework that jointly trains tool creation and tool use inside a single policy. Each rollout is either a **build** task (write a tool from a few examples) or a **use** task (invoke a pooled tool on a held-out question). Three separate reward axes catch schema, code, and outcome failures independently, so each failure mode contributes its own gradient. A 4B Qwen3 trained with SMITH on 13 procedural reasoning tasks with exact verifiers reaches $79.8$ macro-average accuracy on held-out tasks, the best across all evaluated methods and ahead of an untrained 30B-A3B tool-writer. It also reaches $40.4$ on TabMWP-Hard and $42.6$ on out-of-domain GQA ($+7.6$ over the best same-backbone inference-time baseline), without any visual or tabular training data. When invoked by a frozen 350M student, tools written by our 4B match those produced by a writer an order of magnitude larger. The same recipe also lifts Qwen3-8B and Granite-3.3-8B without modification.

Selective knowledge distillation has become increasingly important for compressing large language models, where dense logit-based knowledge transfer over long sequences and large vocabularies incurs substantial computational overhead. Distillation supervision is naturally organized over a sequence–vocabulary matrix, with its highest-information components often concentrated on a small subset of specific sequence–vocabulary pairs. Existing methods typically select sequence positions or vocabulary candidates independently, relying on coarse-grained one-dimensional heuristics that weaken the precision and effectiveness of knowledge transfer. To address this, we propose $\textbf{J}$oint $\textbf{S}$equence–$\textbf{V}$ocabulary $\textbf{D}$istillation ($\textbf{JSVD}$), a plug-and-play selective distillation framework that identifies and focuses supervision on informative sequence–vocabulary pairs. For each training example, JSVD constructs bidirectional teacher--student candidate supports, scores sequence--vocabulary pairs using prediction discrepancies, and progressively focuses distillation on unresolved informative pairs during training. Extensive experiments across diverse distillation objectives, model pairs, and benchmarks show that JSVD consistently improves downstream performance while reducing distillation overhead and accelerating convergence.

In complex embodied visual reasoning scenarios, an agent often has only a limited field of view, and the evidence needed to answer a question may be distributed across time, viewpoint, and interacting objects. A model may therefore give a plausible answer without ever observing the relevant object, time, or view that supports it. Current visual reasoning benchmarks largely evaluate passive observations and final answers, overlooking settings that require active perception and evidence acquisition. We introduce JRDB-AVR, a benchmark built from real-world human-scene robotics data that turns this gap into an explicit evaluation: a system receives a visual reasoning question, requests bounded observations by timestamp and viewing angle, and is evaluated on both the final answer and the grounded visual evidence supporting it. The benchmark contains diverse questions over multiple real-world test environments involving temporal search, viewpoint selection, and human-oriented compositional reasoning. We also introduce JRDB-AVR-Agent, a reference active reasoning method that maintains an explicit observation-grounded graph-based world model and answers through solving. Experiments reveal a substantial gap between answer accuracy and evidence accuracy in current baselines, showing that current VLMs can produce unsupported correct answers and that active evidence-aware evaluation is necessary for embodied visual reasoning.

Network interference complicates A/B testing on online platforms, such as social networks and marketplaces, where causal methods based on single experiments often suffer from significant bias due to complex interference patterns. This paper demonstrates the statistical benefits of merging data from multiple experiments with varying treatment proportions. Sequential experimentation with increasing traffic, or ramp-up, is widely used in tech companies for risk management and cost control. Beyond operational benefits, we show that regression-based estimators trained on merged data achieve substantial bias reduction, even under simple randomization schemes and regression models. We focus on the global average treatment effect (GATE), a key estimand in the tech industry, and consider a general interference pattern that extends beyond the 1-hop setting. We present an exact bias–variance analysis of the linear regression estimator and show that, in practical settings, the bias term typically dominates. In addition, we characterize the substantial bias reduction achieved by our proposed approach, which merges experimental data collected at different ramp-up stages to improve the training of the regression function. We also offer an intuitive explanation for this reduction and highlight the synergy between cluster-level randomization and our approach. Furthermore, we consider a refined estimator based on graph neural networks (GNN). Extensive simulations across novel and challenging scenarios confirm that our methodology significantly improves the accuracy of regression-based estimators.

LLMs are increasingly applied to cybersecurity workflows, where they are expected to translate analysts' intent into tool invocations. However, existing evaluations focus on knowledge-based assessments or end-to-end agentic tasks, and do not directly measure LLMs’ ability to generate executable commands for real-world cybersecurity tools. This gap is critical because cybersecurity operations rely on strict command-line interfaces (CLIs), where minor syntax errors, incorrect flag-value bindings, or argument misordering can invalidate execution. We introduce KaliBench, a fine-grained benchmark and dataset for natural-language-to-CLI translation on Kali Linux, comprising 8,504 query-command pairs spanning 1,642 tools across 23 capability dimensions and 5 security phases. KaliBench is constructed via a manuscript-grounded pipeline with deterministic canonicalization and alias-aware evaluation, enabling precise and reproducible assessment of tool selection and argument construction. To ensure both semantic correctness and practical executability, we develop a multi-stage verification pipeline that combines LLM-based validation, sandboxed terminal execution, and human-in-the-loop refinement. Building on these fine-grained, deterministic signals, KaliBench further enables runtime-free verifiable rewards for training. Across three evaluation modes and over 20 general-purpose and security-focused LLMs, no model exceeds 35\% exact-command accuracy in the unrestricted setting, highlighting the difficulty of accurate CLI-based cybersecurity tool use without explicit tool hints. We further show that supervised fine-tuning and reinforcement learning with verifiable rewards derived from KaliBench significantly improve an 8B model and achieve performance comparable to a 685B MoE model. KaliBench will be publicly released.

Large language models (LLMs) are increasingly used for knowledge base question answering (KBQA), where answering requires selecting entities from a question-specific knowledge-graph subgraph. Yet LLMs are known to hallucinate across tasks, and KBQA is no exception: even when we provide a graph as the knowledge source, the model may rely on parametric knowledge instead of graph evidence or perform invalid reasoning over the given relations. Such hallucinated answer nodes can limit the practical deployment of KBQA systems, especially in high-stakes domains such as healthcare. We formulate hallucination detection in KBQA as an answer-node classification problem and propose a lightweight graph-based framework that treats the answering LLM as a black box. KG-Guard represents each KBQA instance as an augmented graph. It initializes node features with semantic representations of KG entities, marks topic entities and LLM-proposed answer nodes with learned vectors, and connect a virtual question node to the topic entities. A graph encoder then produces verification-oriented node representations, and a small MLP classifies each proposed answer node using its graph representation together with the question embedding. Experiments on WebQSP, ComplexWebQuestions, and PUGG show that our detector achieves the highest F1 on all three benchmarks ($82.0$, $87.4$, and $84.3$), outperforming LLM-judge and sampling-based baselines, while having $\sim305\times$ fewer parameters than the reference approaches. Beyond detection, the node-level feedback is actionable: when flagged answers are fed back to the KBQA system for iterative refinement, downstream KBQA F1 improves by $13.0$-$14.5$ points and Exact Match by $16.9$-$17.6$ points.

Modern LLMs continue to exhibit significant variance in behavior across languages, such as being able to recall factual information in some languages but not others. While typically studied as a problem to be mitigated, in this work, we propose leveraging this cross-lingual inconsistency as a tool for interpretability in mixture-of-experts (MoE) LLMs. Our knowledge localization framework contrasts routing for sets of languages where the model correctly recalls information from languages where it fails. This allows us to isolate model components that play a functional role in answering about a piece of knowledge. Our method proceeds in two stages: (1) querying the model with difficult factual questions across a diverse set of languages to generate "success" and "failure" activation buckets and then (2) applying a statistical contrastive analysis to the MoE router logits to identify experts important for knowledge. To validate the necessity of this small number of experts for answering a knowledge question, we deactivate them and re-ask the question. We find that despite only deactivating about 20 out of 6000 experts, the model no longer answers correctly in over 40% of cases. Generally, this method provides a realistic and scalable localization approach to increasingly complex LLMs, but also suggests redundant and highly dispersed knowledge parameterization in MoEs.

Recommender systems, which retrieve items of interest to users based on their interaction histories, have recently witnessed several key technological advancements, benefiting from the semantic representation space and model-data scaling. Nevertheless, practitioners and researchers have frequently encountered critical inconsistencies between simple offline evaluations and complex online results, and this gap still exists in the era of large models. In this work, we address several key factors that contribute to this inconsistency: 1) the large-scale continual learning challenge that assumes a practical demand to maintain long-term model effectiveness under dynamic item pools and distributional shifts; 2) the combinatorial diversity challenge that requires the generation strategy to solve the accuracy-diversity trade-off in both the finer token level and the holistic item level; and 3) the multi-task reward balancing challenge, which is particularly critical for modern generative models that can accurately model the sequence distribution but are less controllable for the multi-task demands. To address these challenges and further pave the way for the development of more realistic recommendation solutions, we introduce KuaiRecV2, a deliberately constructed benchmark with industrial-level datasets from Kuaishou. Specifically, we provide a large-scale interaction dataset with million-scale multi-modal item information and long-term interaction records across 30 days, with a dynamic video candidate pool. Additionally, we provide a rigorous benchmark that evaluates retrieval accuracy, diversity, and multi-task performance in one rubric system, where an effective advancement occurs if and only if the new method improves all metrics. The code and data are available at Supplementary Materials.

Improving compositional understanding in CLIP-style vision-language models remains difficult, models with strong coarse alignment often fail on attribute binding, relations, and multi-object semantics. A common remedy is text-side hard negatives (negations, perturbations, or mined confusions), but concentrating difficulty on the text branch can skew the training signal, exacerbate modality asymmetry, and distort joint-space geometry (e.g., increased hubness and reduced mutual reciprocity), even when Recall@K improves. We propose LACE (Latent Alignment via Counterfactual Embeddings), a lightweight approach that injects structured, semantics-preserving latent edits into the image and text embedding streams without pixel-space editing or external generators. LACE synthesizes counterfactual image and text embeddings by attenuating, swapping, or transplanting a small set of factors tied to target objects, attributes, or relations, producing image-side and text-side hard negatives that rebalance supervision while preserving global alignment. Across compositional and retrieval benchmarks, LACE improves robustness to binding and relational confusions and yields a healthier embedding geometry with improved reciprocity and competitive hubness/local-stability trade-offs.

Reinforcement learning (RL) has become central to post-training large language models, but its rollout generation stage is often the dominant system bottleneck. The inefficiency comes from two sources. First, autoregressive decoding makes latency grow with response length. Second, response lengths vary widely within a batch: short sequences finish early, but the batch remains blocked by the longest sequence, leaving GPU capacity underutilized. This effect is amplified in multi-sample rollouts, where long responses can also concentrate on a few decoding workers and delay synchronization. We propose LADDERS (Length-Aware Data Distribution and Existing-Response Speculation), a lightweight framework for accelerating RL rollout generation. LADDERS first uses hidden-state-based length prediction to group prompts with similar expected response lengths, and then applies an S-shaped allocation rule to balance multi-sample requests across workers. It further reuses responses already generated by the policy as draft continuations through prompt-specific suffix trees, enabling speculative decoding without an auxiliary draft model. Experiments on Qwen3-1.7B and Qwen3-8B show that LADDERS reduces rollout generation time by up to 56% while preserving final task performance, and can be integrated into existing RL systems with minimal engineering effort.

Modern video generators produce visually compelling clips but still struggle with physical and motion consistency, limiting their use as reliable world simulators. Existing remedies often rely on external simulators, teacher models, or curated physics-focused data. We explore a complementary self-supervised direction: extracting motion cues from the unlabeled videos already used to train video diffusion models. We propose LaMo, which formulates a latent motion prior over frame-to-frame latent changes conditioned on the current latent and prompt. This prior is exposed through two lightweight readouts: a macro motion drift used during training as a Motion Drift Loss, and a learned micro motion field used during sampling as Motion Prior Guidance. Both components are plug-and-play with existing video diffusion backbones, requiring no architectural or I/O changes. On VideoPhy and VideoPhy2, LaMo improves CogVideoX backbones and outperforms recent physics-aware baselines that use external supervision. On VBench, it preserves overall generation quality while improving motion-related dimensions. These results suggest that unlabeled video contains useful motion supervision for improving physical fidelity in modern video diffusion models.

Long-context inference in Large Language Models (LLMs) suffers from high latency because attention computation and memory traffic scale linearly during decoding. KV cache pruning reduces this cost by retaining only a subset of cached tokens, but its effectiveness depends on whether the pruning proxy can recover the tokens with the largest exact attention logits under the current query. Existing proxies suffer from ranking misalignment: static heuristics rely on historical statistics and thus ignore the current query, while hashing-based retrieval proxies rank candidates in a discrete metric space that does not coincide with attention's inner product. To avoid these problems, we propose LAPrune, a KV pruning framework that uses additive vector quantization to build an efficient attention-logit-aligned proxy. LAPrune represents each cached key as a sum of learned codewords and decomposes the query-key dot product into lookup-and-add operations over query-codeword inner products. This replaces high-dimensional dot products with lightweight scalar lookups while preserving fidelity to the original attention ranking. LAPrune requires no base-model finetuning and can be integrated into long-context decoding with minimal modification. Across long-context evaluations, LAPrune improves exact-logit Top-$k$ recovery and perplexity over hashing-based pruning proxies, achieving up to $2.73\times$ decoding speedup over FlashAttention-2 and $4.83\times$ end-to-end speedup over the vanilla baseline at 128K context. Our code is available at [Anonymous/LAPrune](https://anonymous.4open.science/r/LAPrune-VQ).

Covariate-adjusted response-adaptive randomization (CARA) designs can improve statistical efficiency and participant welfare in randomized experiments by learning from accrued data and dynamically updating treatment allocation. In practice, two issues limit their impact. Many experiments are sample-size constrained because enrollment is difficult and trials are expensive. At the same time, studies increasingly collect rich pre-treatment information, including unstructured data such as text, images, and clinical notes, yet most CARA designs rely on a small set of structured covariates when computing allocation probabilities. Recent advances in large-scale pre-trained large language models (LLMs) offer a new opportunity to extract useful signals from unstructured inputs and external knowledge. However, few-shot LLM predictions are not ordinary fixed fitted predictions because prompt demonstrations sampled from accrued trial data create data-dependent randomness and can induce correlation across predictions. If used naively to drive allocation updates, AI-generated signals may weaken CARA's ability to improve power and participant welfare. We propose a CARA design that integrates few-shot LLM-predicted outcomes through resampling-based aggregation, calibration, and effective residual variances to guide both adaptive treatment allocation and downstream treatment effect estimation. Theoretical results and simulation studies show that, when calibrated few-shot LLM auxiliary signals reduce residual variation, coupling an LLM-enhanced estimator with an adaptive allocation rule can improve statistical efficiency and participant welfare.

The ability to learn a new task from a few examples has been crucial to the success of foundation models such as large language models (LLMs). However, current foundation models for structured data like tables and relational databases still require tens of thousands of labeled examples to perform well on new tasks. Here we show that when pretrained at scale with the right recipe, Relational Transformers (RTs) can make state-of-the-art (SoTA) predictions with only hundreds of labels. This capability arises from the following ingredients: First, we assemble THE JOIN, the largest pretraining corpus of relational data to date, comprising 6k forecasting tasks across 650 real-world databases from diverse domains. Second, we develop a pretraining recipe that combines (1) mixed context sizes for multi-scale learning, (2) multi-cell masking for dense supervision, and (3) a novel random- walk-based retriever to efficiently gather relevant context. Third, we identify context ensembling and context tuning as complementary axes for scaling test-time compute to further improve few-shot performance. Pretrained with our recipe on THE JOIN, RT achieves parity with LLM Agent + TabICLv2 and RDBLearn + TabICLv2 pipelines with 32–23×fewer labels respectively, and even surpasses the prior SoTA of full task-specific training, on average nMAE for RelBench regression tasks. Context ensembling and tuning improves this further by up to 3%, and 4% respectively. Our ablations highlight the importance of schema semantics, multi-cell masking, and random-walk retrieval. Overall, our work paves the way for developing relational foundation models with strong few-shot capabilities.

In this paper, we study last-iterate convergence in stochastic constrained convex--concave minimax optimization. A key difficulty is that the last iterates of vanilla stochastic extragradient (S-EG) and stochastic optimistic gradient descent--ascent (S-OGDA) can fail to converge in the presence of gradient noise, even for simple bilinear problems. To address this issue, we regularize the original convex--concave problem into a strongly convex--strongly concave one. Applying S-EG and S-OGDA to the regularized problem gives two simple single-loop first-order methods, which we call perturbed S-EG and perturbed S-OGDA. By carefully choosing the regularization parameter and balancing the resulting regularization bias, stochastic error, and optimization error, we prove that both methods achieve a last-iterate convergence rate of $\mathcal{O}(T^{-1/4+\varepsilon})$ for any $\varepsilon>0$ in terms of the primal--dual gap. This improves upon the best-known $\tilde{\mathcal{O}}(T^{-1/7})$ rate under comparable constrained settings. Moreover, for unconstrained problems, we establish almost sure convergence under a single-timescale stepsize scheme, whereas prior almost sure convergence results typically rely on two-timescale stepsizes and are limited to S-EG.

We study learning-augmented online portfolio selection (OPS), where an investor uses predictions to improve wealth while remaining protected against unreliable advice. In frictionless markets, we characterize the optimal tradeoff between robustness (wealth under arbitrary predictions) and consistency (wealth under perfect predictions) tied to the geometric mean, and show that the optimal tradeoff admits a tractable online characterization. We further quantify how performance interpolates between these endpoints under imperfect predictions by smoothness guarantees. Under transaction costs, we give exact robust-trading certificates and, to our knowledge, the first learning-augmented OPS impossibility under costs: for any positive cost, no GM-robust policy can be maximally consistent under exact predictions. Lastly, we show that a cost-aware greedy rule remains Pareto-optimal.

Mobile sensing for real-time anomaly detection couples routing, sampling, and stopping: dispatch determines which locations are observed, while observations update the evidence used for future dispatch. We study this problem for a fleet of unmanned aerial vehicles (UAVs), where each UAV observes only its current location and anomalies may occur at unknown locations and times. The goal is to detect anomalies quickly while controlling false alarms under decision-dependent partial observations, local mobility, and collision-avoidance constraints. Existing quickest-detection methods typically ignore mobility and collision constraints, whereas learning-based routing methods scale to large fleets but are not designed for statistically calibrated detection. We formulate route-wise monitoring as a partially observable Markov decision process (POMDP) and propose DispatchPPO, a deep reinforcement learning method for long-horizon detection-aware dispatch. Its policy architecture, DispatchNet, uses autoregressive decoding with dynamic feasibility masking to generate collision-free joint UAV moves without enumerating the exponential joint action space. We establish theoretical properties for persistent coverage, evidence-based focusing, and detection delay through a resource-allocation information-rate bound. Experiments show that DispatchPPO reduces detection delay relative to heuristic and optimization-based baselines, while transferring effectively to real-robot and wildfire monitoring case studies.

Efficient exploration in sparse-reward environments, where informative extrinsic feedback is scarce, remains a significant challenge in reinforcement learning. A key limitation of existing exploration strategies is that they often assume densely coupled environment dynamics, overlooking the underlying causal mechanisms of the environment, particularly the fact that causal relationships may vary across contexts. To address this issue, we propose contextual causal dynamics model-based intrinsic motivation (CIM), a causality-aware exploration framework that explicitly captures context-dependent sparse causal structures and derives intrinsic motivation signals, specifically causal action influence and curiosity. These intrinsic rewards encourage interventions on causally relevant state factors rather than merely promoting observation novelty-seeking behavior, thereby facilitating robust exploration and active skill acquisition. Moreover, by learning contextual causal mechanisms, CIM identifies and filters out task-irrelevant state variables, improving generalization under distribution shifts. Empirical results demonstrate that our method achieves superior sample efficiency, training robustness, and out-of-distribution generalization performance across multiple robotic manipulation tasks.

Prefill-Decode (PD) disaggregation improves LLM serving by isolating compute-bound prefilling and memory-bound decoding onto specialized GPU pools. However, dynamically rebalancing these pools is challenging because request arrival rates, prompt lengths, and generation lengths change over time. We formulate this autoscaling problem as latency-constrained cost minimization: the controller must minimize GPU usage while keeping P99 tail latency within TTFT and TPOT targets. Existing threshold-based heuristics require extensive parameter tuning, yet still tend to over-provision. To address this, we propose a reinforcement learning (RL) framework that learns a cost-efficient autoscaling policy. The policy uses a recurrent network to track multi-timescale workload trends and pod lifecycle states, enabling decisions that account for delayed pod readiness and delayed SLO feedback. It also restricts the action space with hard masks and anti-oscillation cooldowns, enforcing feasible scaling actions while reducing wasteful GPU transitions. We train the policy in a high-fidelity PD-disaggregated serving simulator calibrated from profiling data and validated against a physical testbed. Simulator experiments on ShareGPT and Azure Conversational traces show that the learned policy satisfies the same latency constraints as grid-searched heuristic baselines while reducing average GPU occupancy. Real-cluster deployment on ShareGPT further confirms sim-to-real transfer while satisfying the latency targets.

Learning an energy-based model from data samples is a central problem in machine learning. Many recent and popular methods, such as denoising score matching for training energy-based diffusion models, use stochastic interpolants to corrupt data samples at different noise levels indexed by a time variable. This defines a joint density over both the data space and time, and most methods learn its energy through either spatial or temporal differences. We identify distinct failure modes for both of these approaches. To solve them, we propose Spatiotemporal Noise-Contrastive Estimation (stNCE), a framework for learning the energy through joint spatiotemporal differences. stNCE unifies many existing methods and leads to new training objectives. Experiments on images and molecules demonstrate performance competitive with state-of-the-art density estimation methods.

Reinforcement learning has achieved remarkable success in robot learning. However, under challenging exploration and contact-rich dynamics, early-stage training is frequently dominated by premature terminations such as collisions and falls. As a result, learning is overwhelmed by short-horizon, low-return trajectories, which hinder convergence and limit long-horizon exploration. To alleviate this issue, we propose a technique called Failure Episodic Memory Alert (FEMA). FEMA explicitly stores short-horizon failure experiences through an episodic memory module. During interactions, it retrieves similar failure experiences and prevents the robot from recurrently relapsing into unstable states, guiding the policy toward long-horizon trajectories with greater long-term value. FEMA can be combined easily with model-free reinforcement learning algorithms, and yields a substantial sample-efficiency improvement of 33.11\% on MuJoCo tasks across several classical RL algorithms. Furthermore, integrating FEMA into a parallelized PPO training pipeline demonstrates its effectiveness on a real-world bipedal robot task.

On-policy self-distillation (OPSD) has proven effective for post-training large language models (LLMs), yet its application to diffusion LLMs (dLLMs) remains unexplored. Existing OPSD methods are inherently autoregressive-centric. They inject privileged information via left-to-right prefix conditioning with token-level divergence supervision, a design that fundamentally conflicts with the arbitrary-order generation of dLLMs. We introduce d-OPSD, the first OPSD framework tailored for dLLMs. Our approach makes two core contributions. First, we reframe self-teacher construction by using self-generated answers as suffix conditioning, enabling the student model to learn from ``self future-experience" rather than privileged prefixes. Second, we shift supervision from token-level to step-level, aligning training with the iterative denoising process of dLLMs. Experiments across four reasoning benchmarks show that d-OPSD consistently outperforms RLVR and SFT baselines with superior sample efficiency, requiring less than $10\%$ of the optimization steps by RLVR and opening a promising pathway for dLLM post-training.

Diffusion models have recently emerged as a powerful paradigm for generating 3D molecules. Despite showing remarkable promise, they learn atomic distributions implicitly from data, without explicitly encoding physicochemical rules into their generative dynamics. Consequently, generated samples can be statistically plausible yet chemically invalid. To address this problem, we propose Sequential Neuro-Symbolic Constrained Diffusion (SensDiff), a framework that integrates symbolic rules into generative modeling, enabling the model to internalize scientific priors during training. Central to SensDiff is the observation that the diffusion trajectory of 3D molecule generation exhibits a coarse-to-fine evolution: valid molecules arise by first mitigating macroscopic geometric conflicts and then refining structures toward microscopic chemical consistency through iterative denoising. Accordingly, SensDiff mirrors this dynamic by sequentially enforcing constraints with adaptive weighting during training, progressively steering generation toward valid molecular geometries. Moreover, SensDiff translates domain knowledge into interpretable generative controls, grounding opaque neural denoising in scientific principles. Experiments on molecular benchmarks and real-world constrained generation tasks confirm that SensDiff consistently improves chemical validity while adhering to physicochemical rules and user-specified targets.

Vision Transformers (ViTs) are bottlenecked by the length of their input token sequence. Standard patchification fixes this length with a single patch size applied across the image, regardless of regional content. Prior work on adaptive patch sizing reduces token count by assigning coarser patches to visually simple regions, but relies on noisy pixel-level heuristics such as local entropy to make patchification decisions. We argue that this compute allocation should be grounded in the model’s own semantic representations, not in raw pixel statistics. We introduce Split Policy Learned Image Tokenization which learns to allocate token resolution based on what the model finds meaningful, producing shorter token sequences that are better matched to task-relevant content. SPLIT matches the accuracy of full-resolution ViT classification baselines while using 30% fewer tokens. The same gains hold dense prediction: SPLIT achieves competitive accuracy on COCO and ADE20K, demonstrating that semantically grounded patchification generalises well beyond classification.

Budget-feasible mechanisms (BFMs) are a fundamental tool for budget-constrained procurement, but designing strong BFMs remains challenging, particularly for general valuation classes. Classical approaches rely on problem-specific analysis and worst-case approximations, often yielding conservative mechanisms in practice. This motivates data-driven methods for discovering mechanisms with stronger empirical performance. However, strict budget constraints expose a key limitation of existing neural automated mechanism design (AMD): they either fail to provide exact economic guarantees or suffer from severe scalability bottlenecks. We propose BFMNet, a scalable menu-based neural framework for learning truthful budget-feasible procurement mechanisms. BFMNet first learns *self-bid-independent* menus from market context, ensuring dominant-strategy incentive compatibility (DSIC) and individual rationality (IR) by construction, and then applies an ex-post, instance-level menu transformation to enforce budget feasibility. We prove that the resulting mechanism satisfies exact DSIC, IR, and budget feasibility. By avoiding intractable global discretization grids and restrictive Lipschitz constraints, BFMNet remains both expressive and scalable. Extensive experiments show that BFMNet improves utility over classical baselines by up to $34.1$%, remains competitive with state-of-the-art neural baselines that provide only approximate economic guarantees, and scales robustly to larger auction instances.

Optimal Transport (OT) provides a principled framework for comparing probability distributions. Classical solvers remain the gold standard for this task, providing exact solutions. At the same time, OT is an ideal testbed for studying whether learning-based amortized models can internalize the structure of OT type problems. Here, ground-truth solutions are available, constraints are explicit, and exact discrete solutions have well-characterized sparsity. We describe ToR-OT, a transformer-based model that learns to predict discrete OT plans through autoregressive token regression. We reframe OT into a structured sequence generation task. ToR-OT decodes sparse transport plans as sequences rather than solving an optimization task at test time. We focus on three properties: (a) Zero-shot generalization: pretraining on a large synthetic corpus of OT instances allows a single model to generalize to unseen distribution pairs and heterogeneous cost metrics without retraining. (b) Sparse primal decoding: by exploiting the inherent sparsity in exact discrete OT solutions, the decoder predicts only nonzero entries (avoids dense coupling outputs). (c) Amortized inference: on low-dimensional discrete problems, ToR-OT gives accurate approximate plans and compares favorably with per-instance neural OT baselines. Overall, ToR-OT is not yet a replacement for classical solvers, but best viewed as a step toward understanding how autoregressive transformers can learn structure of constrained optimization problems like OT.

Open-vocabulary dense perception requires grounding language-specified concepts in local regions for recognition, localization, and segmentation beyond closed category sets. CLIP learns image-level vision-language alignment, whereas dense prediction requires spatially precise and semantically reliable local features. Although recent dense CLIP adaptation methods reduce this gap through region-level semantic transfer and spatial correlation guidance, maintaining stable local semantics under contextual variation remains challenging. Specifically, the global contextual modeling in CLIP may introduce undesirable dependencies into local features. Changes in background, scale, viewpoint, or scene layout may cause identical object regions to yield inconsistent dense representations, leading to unstable correspondences and semantic drift. Moreover, spatial correlation guidance from visual foundation models captures visual relatedness between patches, but not whether coherent regions correspond to the queried textual concept, leaving concept-level dense alignment under-constrained. To address these issues, we introduce Semantic Spatial Consistency Learning $\textbf{(SSCL)}$, an adaptation method with two complementary objectives: 1) Context-Invariant Spatiotemporal Consistency (CISC) enforces semantic consistency for the same anchor region under varying contexts to mitigate context-induced feature drift; 2) Structure-Guided Semantic Refinement (SGSR) leverages spatial affinities from a frozen visual foundation model to refine CLIP text-patch responses, producing text-conditioned dense supervision with improved spatial coherence and concept-level alignment. Experiments on video instance segmentation, region classification, semantic segmentation, and object detection show consistent gains over the baseline methods, demonstrating the effectiveness and generalization capability of the proposed SSCL. Our code and models will be made publicly available.

Understanding how complex molecular systems transition between metastable states is a central challenge in computational chemistry and biophysics. A key tool for characterizing these transitions is the committor function, defined as the probability that a trajectory initiated from a given configuration reaches one state before another, which encodes detailed mechanistic information about transition pathways. First, we curate a public benchmark of empirically determined committor values for alanine dipeptide. Second, we develop a "committor-in-the-loop" framework that combines a self consistent loss function with weighted ensemble (WE) simulation: WE adaptive bins are biased by the learned committor, which is in turn trained on WE samples. Across a two-channel 2D system and alanine dipeptide settings, our WE-based framework consistently outperforms greedy baselines, demonstrating the importance of principled exploration. Together, the benchmark and framework provide both a standard for evaluating committor learning methods and a new method for studying rare transitions in high-dimensional molecular systems.

The rapid evolution of Time Series Foundation Models (TSFMs) has advanced zero-shot forecasting across diverse domains. Inspired by the current form of Large Language Models, future TSFMs may be offered as commercialized, closed-source API services. However, many existing online adaptation methods still rely on white-box access for parameter fine-tuning or gradient backpropagation. This paradigm mismatch raises a question: In black-box online adaptation for TSFMs, what should we learn? We answer this with an insight: the predictive errors of the base model are conditioned on both the input and output of the base model (i.e., the context of errors). To validate this insight, we propose ORCA (Online Residual Contextual Adaptation). We conduct extensive experiments across 5 state-of-the-art TSFMs and 8 datasets to demonstrate the effectiveness of our approach. Furthermore, through ablation studies, we quantitatively analyze the impact of different adapter learning hypotheses on the final adaptation performance in black-box online adaptation. Code available at https://anonymous.4open.science/r/ORCA1/.

Edge-cloud collaboration between large language models (LLMs) and small language models (SLMs) offers a promising way to leverage the problem-solving capability of LLMs while keeping private data on device. Existing methods typically have the LLM generate guidance from an initial user query, which the on-device SLM combines with local private data to produce the final response. However, the LLM's general guidance often does not align with the user's personalized needs, limiting the utilization of the LLM's capabilities. It has been observed that the on-device SLM exhibits metacognitive capabilities, which allow it to recognize its limitations and adjust its reasoning process. Motivated by this, we transform the on-device SLM into a metacognitive inquirer agent that can assess its internal state and actively decide when, what, and how to query the LLM for personalized guidance. Specifically, we propose Active Inquiry via Metacognitive Actions (AIMA). AIMA learns an on-device policy that jointly optimizes the selection of high-level metacognitive actions and the action-conditioned query refinement. We further introduce a privacy-constrained rewriting mechanism that detects and eliminates sensitive information leakage in the refined query. Extensive experiments and theoretical analysis demonstrate that our method outperforms existing methods.

Intracortical brain–computer interfaces (iBCIs) can restore movement and communication, but neural drift can degrade performance across recording sessions. Because recalibration adds user burden, there is a need for decoders that adapt from minimal calibration data. Although session-specific input layers are a common way to handle cross-session drift, they can underperform in the extreme few-shot regime because the shared decoder is trained downstream of well-estimated session-specific layers, but must rely at deployment on a new-session layer fit from only a few calibration trials. To avoid this dependence, we propose a two-stage approach: first, train a single GRU decoder on pooled held-in sessions without session-specific input layers, encouraging the backbone to handle cross-session variability directly; second, freeze this backbone and meta-learn a lightweight alignment layer for rapid calibration. We evaluate on FALCON H2, an official human handwriting iBCI benchmark designed for few-shot cross-session decoding, and achieve 7.41\% word error rate using only three released calibration sentences per held-out session. These results suggest that, under extreme calibration limits, learning a transferable backbone before adding session-specific adaptation can be more effective than jointly training flexible session-specific layers.

Complex visual question answering often requires vision-language models to think with images, i.e., actively acquire local visual evidence by zooming into informative regions of high-resolution images. Our error analysis shows that many failures in this setting originate from incorrect observation actions, where the model fails to localize key regions or obtain crops that contain sufficient evidence for answering. We further observe that bounding-box tokens exhibit substantially higher generation entropy than ordinary language tokens, indicating high uncertainty when the model decides where to look. However, existing reinforcement learning methods based on final-answer rewards provide only trajectory-level feedback, making it difficult to assign precise credit to the observation actions. We propose Observation Policy Optimization (OPO), an RL algorithm for optimizing observation actions in tool-augmented thinking with images. OPO consists of two core modules. First, Uncertainty-Guided Observation Branching (UOB) treats each image zoom-in tool call as an observation event and selectively branches at high-uncertainty, low-redundancy events during rollout, enabling efficient exploration of alternative cropping decisions and their resulting visual evidence. Second, Evidence-Localized Advantage Attribution (ELAA) compares sibling branches using local evidence-sufficiency rewards and assigns credit only to the corresponding observation-action tokens. By decoupling observation-action optimization from trajectory-level outcome supervision, OPO improves visual evidence acquisition and further enhances visual reasoning performance. Experiments show that OPO consistently improves Qwen3-VL models across scales and achieves competitive performance against larger models and tool-augmented baselines.

We study the problem of learning with multiple correct answers, where each instance admits a set of valid labels. We primarily focus on the online setup, where in each round the learner must output a valid label for the queried example. This setting is motivated by language generation, in which a prompt may admit many acceptable completions, but not every completion is acceptable. We study this problem under three feedback models. For each model, we characterize the optimal mistake bound in the realizable setting using an appropriate combinatorial dimension. We then show that the rate of regret can be constant, linear, or sublinear across the three models in the agnostic setting. Our results also imply sample complexity bounds for the batch setup that depend on the respective combinatorial dimensions.

Graphical User Interface (GUI) grounding maps natural language instructions to the visual coordinates of target elements and serves as a core capability for autonomous GUI agents. Recent reinforcement learning methods (e.g., GRPO) have achieved strong performance, but they rely on expensive multiple rollouts and suffer from sparse signals on hard samples. These limitations make on-policy self-distillation (OPSD), which provides dense token-level supervision from a single rollout, a promising alternative. However, its applicability to GUI grounding remains unexplored. In this paper, we present GUI-SD, the first OPSD framework tailored for GUI grounding. First, it constructs a visually enriched privileged context for the teacher using a target bounding box and a Gaussian soft mask, providing informative guidance without leaking exact coordinates. Second, it employs entropy-guided distillation, which adaptively weights tokens based on digit significance and teacher confidence, concentrating optimization on the most impactful and reliable positions. Extensive experiments on six representative GUI grounding benchmarks show that GUI-SD consistently outperforms GRPO-based methods and naive OPSD in both accuracy and training efficiency. Code and training data will be publicly released.

Large language model agents now combine tools, long context, memory, and multi-step workflows, yet they still fail when execution loses track of the evidence required for a reliable answer. Common failure modes include missing support, unresolved contradictions, and evidence lost under context pressure. We present Lemon, an evidence-risk-aware framework that makes such risks an explicit control variable. At the intra-instance level, a single state-conditioned controller jointly schedules reasoning, tool invocation, worker expansion, verification, context compression, and memory operations. A recoverable evidence substrate pairs anchored context compression with reusable semantic memory, preserving raw tool outputs while storing transferable process fragments. We further extend Lemon to an inter-instance setting in which heterogeneous personalized agents expose expertise, exchange evidence-bearing proposals, critique unsupported claims, and internalize useful collaboration artifacts. Empirically, Lemon reaches 91.36\% accuracy on GAIA and 80\% on xbench-DeepSearch. In an open-source reproducibility comparison on GAIA, Lemon uses 49.9\% to 88.4\% fewer tokens per task than three top-ranked agent baselines.

Current skeleton representation learning paradigms face distinct limitations: Contrastive Learning (CL) often overlooks fine-grained motion details, while Masked Auto-Encoders (MAE) rely on coordinate-level reconstruction. This reconstruction inherently demands dense token sequences and heavy decoders, wasting pre-training computation on discarded components and forcing downstream inference to process dense token grids. To resolve these bottlenecks, we propose SLiM (Skeleton Less is More), a compact-token framework that unifies masked feature prediction and contrastive learning via a shared encoder. By shifting the objective from raw coordinate reconstruction to decoder-free, teacher-guided feature prediction, SLiM breaks the reliance on dense tokenization and enables effective learning with a highly compact token grid. Crucially, to prevent trivial shortcut learning arising from strong inter-joint dependencies of human, we introduce Semantic Tube Masking together with Skeleton-Aware Augmentations to enforce deep skeletal-temporal reasoning and anatomical consistency. Extensive experiments across multiple downstream protocols demonstrate that SLiM achieves state-of-the-art performance while structurally reducing inference computation by 7.89$\times$ compared to dense-token MAE baselines.

The performance of machine learning models often relies on large labeled datasets; however, data collected from diverse sources can contain label noise. Recent work has shown that, in noisy settings, there may exist a subset of the training data on which models can achieve performance comparable to training on a noise-free dataset. A widely used method for identifying such subsets is \textit{cutstats}, which employs $k$-nearest neighbors ($k$-NN) to detect low-noise samples. However, its performance on high-dimensional data remains largely unexplored. In this work, we formally establish that the performance of a classifier trained on a subset of a noisy dataset selected via \textit{cutstats} is influenced by the accuracy of $k$-NN. We further demonstrate that, in noisy environments, exploiting data invariance and knowledge of underlying symmetries can significantly enhance the performance of $k$-NN, bringing it closer to the Bayes optimal classifier even in high-dimensional regimes. Finally, we show that for real-world scenarios, where information about the underlying invariance is only partially known, learnt invariant representations can still facilitate the identification of near-optimal subsets.

We introduce a novel Longitudinal Focused Attention Meditation Electroencephalography (L-FAME) dataset and an accompanying benchmark, designed to foster research into the neural effects of various meditation practices and the evolution of these effects over a six-week training period. The dataset contains EEG recordings and psychological assessments from 74 healthy college participants, collected at two distinct time points: pre-intervention and post-intervention. Participants were randomly assigned to one of three distinct meditation groups: two mantra-based techniques (SA-TA-NA-MA and Hare Krishna) and one Breath Focus practice. Leveraging this unique longitudinal and comparative dataset, we propose a benchmark suite comprising three distinct classification tasks: (1) cognitive state decoding to distinguish between resting and meditation states, (2) fine-grained classification of the specific meditation techniques, and (3) cross-session adaptation to evaluate model generalization across the longitudinal time gap. We provide comprehensive baseline results for these tasks utilizing a range of classical machine learning algorithms and deep learning architectures. The complete dataset, preprocessing pipelines, and benchmark evaluation code will be publicly released, offering a valuable resource and a standardized framework for the development and comparison of new analytical methods in computational meditation research and EEG-based machine learning.

As AI agents move from chat interfaces to systems that read private data, call tools, and execute multi-step workflows, guardrails become a last line of defense against concrete deployment harms. In these settings, guardrail failures are no longer merely answer-quality errors: they can leak secrets, authorize unsafe actions, or block legitimate work. The hardest failures are often contextual: whether an action is acceptable depends on local privacy norms, organizational policies, and user expectations that resist pre-deployment specification. This creates a practical gap: guardrails must adapt to their own operating environments, yet deployment feedback is typically limited to sparse, noisy user-reported failures, and repeated fine-tuning is often impractical. To address this gap, we propose LiSA (Lifelong Safety Adaptation), a conservative policy induction framework that improves a fixed base guardrail through structured memory. LiSA converts occasional failures into reusable policy abstractions so that sparse reports can generalize beyond individual cases, adds conflict-aware local rules to prevent overgeneralization in mixed-label contexts, and applies evidence-aware confidence gating via a posterior lower bound, so that memory reuse scales with accumulated evidence rather than empirical accuracy alone. Across PrivacyLens+, ConFaide+, and AgentHarm, LiSA consistently outperforms strong memory-based baselines under sparse feedback, remains robust under noisy user feedback even at 20\% label-flip rates, and pushes the latency--performance frontier beyond backbone model scaling. Ultimately, LiSA offers a practical path to secure AI agents against the unpredictable long tail of real-world edge risks.

Outdoor robot navigation in dynamic, unknown environments remains a formidable challenge, necessitating robust traversability assessment and collision-free planning. While traditional map-based approaches struggle with adaptability to novel scenes, existing map-free methods frequently incur high computational costs and depend heavily on LiDAR data. In response, we propose LiteNav, a vision-only for exteroceptive sensing and strictly map-free outdoor navigation system. Relying solely on a single off-the-shelf RGB camera and GPS for goal specification, LiteNav estimates traversability, encodes navigation goals, and generates candidate trajectories via a diffusion-based joint encoding model. These trajectories are then refined by a goal-oriented planner to ensure correctness, efficiency, and safety. Remarkably, LiteNav achieves performance comparable to LiDAR-equipped methods without ever constructing or referencing explicit maps. Deployed on a low-power NVIDIA Jetson Xavier NX, it operates in real time, consuming only one-fifth of the computational resources required by prior approaches. Experiments demonstrate that LiteNav achieves state-of-the-art results on the GND dataset and outperforms existing methods in real-world scenarios. In summary, LiteNav is a map-free outdoor navigation system that operates using monocular RGB perception and GPS goal cues on an embedded platform, demonstrating strong potential for scalable deployment.

Decision trees are prized for their interpretability and strong performance on tabular data, but popular greedy top-down induction algorithms can yield suboptimal and unnecessarily complex structures. Optimal decision tree methods address this through global optimization, yet remain restricted to axis-aligned threshold splits, which limit the expressivity of each node and often force deep, complex trees to capture non-linear feature effects. Shape Generalized Trees (SGTs) generalize threshold splits to learnable univariate shape functions, improving expressivity and enabling more compact trees. However, existing SGT induction algorithms are greedy and offer no optimality guarantees. In this work, we introduce Literati, the first algorithm for optimal SGT induction. We propose a novel AND/OR graph formulation of the problem that jointly optimizes tree structure and shape function complexity. To solve this AND/OR graph, we develop an AO*-based algorithm with two enhancements that improve anytime performance while preserving optimality: a secondary heuristic for OR-node selection and a round-robin policy for AND-node exploration. Across 24 real-world datasets, Literati achieves higher training and test accuracy than state-of-the-art tree approaches.

LLM-enabled applications are rapidly reshaping the software ecosystem by using large language models as core reasoning components for complex task execution. This paradigm shift, however, introduces fundamentally new reliability challenges and significantly expands the security attack surface, due to the non-deterministic, learning-driven, and difficult-to-verify nature of LLM behavior. In light of these emerging and unavoidable safety challenges, we argue that such risks should be treated as expected operational conditions rather than exceptional events, necessitating a dedicated incident-response perspective. Consequently, the primary barrier to trustworthy deployment is not further improving model capability but establishing systematic threat monitoring mechanisms that can detect and contextualize security-relevant anomalies after deployment---an aspect largely underexplored beyond testing or guardrail-based defenses. Accordingly, this position paper advocates systematic and comprehensive monitoring of security threats in LLM-enabled applications as a prerequisite for reliable operation and a foundation for dedicated incident-response frameworks.

Sign Language Video Generation (SLVG) aims to generate realistic and motion-accurate sign language videos from spoken text. Most existing studies heavily rely on intermediate representations (\eg poses or 3D meshes), and adopt a two-stage generation paradigm, where the text is first converted into poses or other visual modalities, which are then used as conditions to synthesize sign language video frames via diffusion models (\eg pose-to-image models). However, the end-to-end SLVG task remains largely unexplored. In this paper, we demonstrate that end-to-end SLVG is feasible without relying on intermediate representations(\eg poses), and we argue that a key challenge lies in how to generate effective and temporally consistent conditions. With well-designed condition modules, it becomes feasible to build an end-to-end SLVG model without relying on any intermediate modalities. To this end, we propose LLMCond, a first end-to-end text-to-video SLVG framework that requires no auxiliary visual cues during either training or inference. LLMCond consists of four key components: (1) a video VQVAE that compresses sign language videos into a sequence of discrete latent tokens; (2) \textbf{an LLM-based conditioner} that generates temporally and semantically aligned conditions corresponding to the sign sequences; (3) \textbf{a Gaussian Condition Refiner} designed to enforce temporal smoothness across conditions, enabling the generation of coherent and natural sign language motions; and (4) a discrete diffusion model that synthesizes motion-accurate sign language videos conditioned on the refined condition sequences. Extensive experiments on public SL datasets demonstrate that LLMCond achieves highly competitive performance, producing temporally coherent and accurate sign language motions without relying on auxiliary visual modalities such as pose, depth, or optical flow.

We investigate the automated generation of executable robot reinforcement learning (RL) policies from natural-language task descriptions. Rather than using large language models (LLMs) as execution-time decision makers, we use them as designers that construct RL pipelines before deployment, including Markov decision process (MDP) formulation and simulator implementation. We identify a phase-wise structure in this problem: MDP formulation is dependency-structured and sequential, whereas simulator implementation becomes modular once the formulation is fixed. Based on this structure, we propose ARPG, a dependency-aware agentic framework that performs sequential MDP modeling, localized verification and correction, and simulator-structured implementation while externalizing intermediate artifacts. ARPG targets a key failure mode of automated policy generation: executable code may still yield non-learnable policies when state, action, reward, and termination artifacts are not semantically aligned. Experiments in Isaac Sim show that ARPG improves modeling correctness, environment executability, and policy-generation success compared with monolithic LLM baselines and off-the-shelf agentic systems. Additional demonstrations on custom robotics, standard RL, and engineering-domain MDPs further suggest that the same modeling pipeline can be reused beyond the main benchmark suite.

Multimodal large language models (MLLMs) have made rapid progress in general visual understanding, yet they still struggle with fine-grained instance-level consistency verification, where the goal is to determine whether two images depict the exact same physical object under varying viewpoints, illumination, and backgrounds. This task requires both a global understanding of the target instance and a precise examination of subtle local evidence, since semantically similar objects may differ only in small discriminative regions. To address this challenge, we propose \textbf{LocalAgent}, a collaborative agentic framework that integrates the complementary strengths of large and small models. Specifically, the MLLM first reasons from a global instance-level perspective to identify suspicious local regions that may determine the final verification result. These accurately localized regions are then delegated to a lightweight Local Evidence Verifier (LEV) toolkit for fine-grained appearance and geometric matching. The LEV toolkit consists of \textbf{Local Appearance Consistency (LAC)}, which provides illumination- and color-robust similarity estimation, and \textbf{Local Geometric Correspondence (LGC)}, which establishes reliable spatial correspondence under viewpoint changes. Rather than directly replacing the MLLM's reasoning, LEV returns structured similarity scores and comparative evidence in a form that the MLLM can effectively interpret for final decision-making. This design enables a coordinated large-small model collaboration: the MLLM contributes global semantic reasoning and region-level suspicion localization, while specialized local verifiers provide precise, invariant, and interpretable evidence for ambiguous cases. We further introduce an answer-dominant reinforcement fine-tuning (RFT) strategy that enables the policy to learn when verification is necessary. Experiments show that LocalAgent achieves a \textbf{+4.8\%} average improvement over direct-inference baselines, effectively narrowing the gap with closed-source models, while preserving the general multimodal capabilities of the underlying MLLM.

Diffusion models are prone to generating structural hallucinations - samples that match the statistical properties of the training data yet defy underlying structural rules, resulting in anomalies like hands with more than five fingers. Recent research studied this failure mode from several viewpoints, offering partial explanations to their occurrence, such as mode interpolation. In this work, we propose a complementary perspective that treats hallucinations as instabilities on the model-induced manifold. We begin by showing that a hallucination filter based on such instabilities matches or exceeds the performance of the recently proposed temporal one. By tracing the source of these instabilities, we identify local intrinsic dimension (LID) as their primary driver and propose Intrinsic Quenching (IQ), a direct corrective mechanism that deflates it to alleviate hallucinations. IQ consistently outperforms standard hallucination reduction baselines across a wide array of benchmarks and offers a highly promising solution for enforcing anatomical consistency in downstream medical imaging tasks.

A kernel $k(x,y)$ is *LSHable* if there exists a locality sensitive hashing scheme $\mathcal H$ such that $k(x,y)=\Pr_{h\sim\mathcal H}[h(x)=h(y)]$ for all $x,y$. This notion plays a key role in efficient kernel methods in high dimensions. In this work, we show that the $p$-exponential kernel $k(x,y)=\exp(-\lVert x-y \rVert_p)$ is LSHable in bounded regions for all $1

Vision-Language Models (VLMs) are increasingly adopted as end-to-end driving planners, yet they suffer from cross-city domain shift. Existing methods treat this shift as a single quantity addressed by uniform alignment or single-site editing. We refute this view on two counts: ***(i) domain shift in VLM trajectory planners is inherently structured across modality and layer, and (ii) a locate-then-repair recipe matched to this structure outperforms every uniform-objective baseline***. We analyze domain shift via activation patching and uncover two regularities. **(F1) Modality axis.** Image and text tokens exhibit distinct shift patterns in each VLM. **(F2) Layer axis.** The layer-wise shift partition is architecture-determined and dataset-stable. These findings recast domain shift as a **two-dimensional tensor** indexed by *modality* (image vs. text tokens) and *layer*, motivating a simple principle: ***locate where shift occurs, and repair only there***. We instantiate this principle in **MoLaRx** (**Mo**dality–**La**yer **R**epair), a locate-then-repair framework. The Causal Importance Score (CIS) estimates the domain shift as a modality–layer tensor and partitions the layers into functional zones. CIS-Guided Selective Repair (CISR) then applies a matched operator (Maximum Mean Discrepancy (MMD) / low-rank adaptation (LoRA) / freezing) to repair each zone, with the composition adapting automatically to each architecture. We evaluate **MoLaRx** on three architecturally distinct VLMs (LFM2-VL-1.6B, Qwen2-VL-2B, and PaliGemma2-3B) across two cross-city benchmarks (nuScenes and Argoverse 2). MoLaRx attains the lowest cross-city trajectory-error gap on every (VLM, dataset) cell, while using $50$–$60\%$ fewer trainable parameters and $8.4$–$10.9$ ms lower inference latency. In contrast, the strongest scalar baseline (MMD-only) collapses on every cell—direct evidence that no scalar objective can repair structured shift. Code: [anonymous.4open.science/r/MoLaRx-E3B4](https://anonymous.4open.science/r/MoLaRx-E3B4/).

Transformers can perform in-context learning at depths far smaller than the number of optimizer updates suggested by existing gradient-descent interpretations. Prior work shows that an L-layer Transformer can emulate L gradient-descent updates, but this leaves open whether depth is merely a step counter. We show that it is not: in linear self-attention, each layer compounds the algebraic capacity built by previous layers, yielding a tight Θ(log k) depth characterization for computing the final result of k gradient-descent steps on diagonal quadratic tasks. Controlled experiments validate this mechanism: trained linear-attention models exhibit the predicted logarithmic depth transition, and kernel and oracle comparisons identify polynomial degree as the governing bottleneck. Broader experiments show the same depth signal under standard Transformer components, broader ICL regression families, and early-exit probes on Qwen2.5.