Diffusion models provide powerful priors for zero-shot video inverse problems, but their real-time deployment is hindered by two inefficiencies: high initial latency caused by holistic video restoration, and low throughput resulting from multiple VAE passes to enforce measurement consistency in pixel space. To overcome these limitations, we propose Autoregressive Video Inverse problem Solver (AVIS). The AVIS framework leverages autoregressive video diffusion models to restore videos in a streaming manner, naturally eliminating latency bottlenecks. Specifically, AVIS initializes reverse diffusion with a measurement-consistent estimate, reducing the required sampling steps. Compared to leading non-autoregressive solvers, AVIS drastically reduces initial latency from 114s to 4s and increases throughput from 0.71 to 1.18 FPS while achieving superior restoration quality. We further introduce a highly accelerated variant, dubbed AVIS Flash, that enforces measurement consistency solely on the first chunk. AVIS Flash substantially boosts throughput to 5.91 FPS on a single RTX 4090 GPU while maintaining competitive performance and achieving a favorable efficiency–performance trade-off, paving the way toward real-time deployment.

This paper addresses the challenge of obtaining strong optimality guarantees in constrained nonsmooth nonconvex optimization under mild regularity conditions, namely local Lipschitz continuity and existence and continuity of directional derivatives. While standard methods typically ensure weak stationarity notions, achieving directional (d-)stationarity remains nontrivial. We show that a random direction exploration step is sufficient to attain d-stationarity. The proposed approach augments any base optimization method with a single exploration step that samples a direction and step size and accepts the candidate based on a function value comparison. The resulting scheme guarantees that all accumulation points are d-stationary almost surely, independently of the behavior of the underlying method. Moreover, it preserves convergence rates of the base method, as established for DCA and prox-linear-type schemes. The theoretical results are complemented by numerical experiments illustrating the effect and guarantees of the exploration step.

Dynamic Scene Graph Generation (DSGG) aims to capture objects and their evolving relations in videos. Despite recent progress, the practicality and quality of generated scene graphs remain limited compared to the rapid advances in Multimodal Large Language Models (MLLMs). In this work, we revisit DSGG from two fundamental perspectives: task setup and model design. From the task setup perspective, we identify two key limitations of the current recall-oriented evaluation protocol: (i) a severe precision–recall trade-off, and (ii) uninformative and redundant relation generation. To better assess practical usefulness, we introduce four additional metrics that provide a more comprehensive evaluation of the quality of generated dynamic scene graphs. From the model design perspective, we explore directly using MLLMs for scene graph generation and establish a strong MLLM-based DSGG baseline through three design changes. First, we replace the conventional bottom-up pipeline with a top-down reason-then-locate strategy. Second, we reformulate frame-wise dynamic graphs as Temporal Relation Set (TRS) prediction, improving both efficiency and performance. Third, we introduce Importance-Aware Finetuning (IAF) to encourage more relevant and diverse relation generation. Extensive experiments on Action Genome, VidVRD, and PVSG show that our approach consistently achieves state-of-the-art performance across multiple metrics.

Activation functions play a central role in neural networks by shaping internal representations. Recently, learning binary activation representations has attracted significant attention due to their advantages in computational and memory efficiency, as well as interpretability. However, training neural networks with Heaviside activations remains challenging, as their non-differentiability obstructs standard gradient-based optimization. In this paper, we propose Heavy-Tailed Activation Function (HTAF), a smooth approximation to the Heaviside function that enables stable training with gradient-based optimization. We construct HTAF as a sigmoid–hyperbolic tangent composite function and theoretically show that it maintains a large gradient mass around zero inputs while exhibiting slower gradient decay in the tail regions. We show that Spiking Neural Networks, Binary Neural Networks and Deep Heaviside neural Networks can be trained stably using HTAF with gradient-based optimization. Finally, we introduce Implicit Concept Bottleneck Models (ICBM), an interpretable image model that leverages HTAF to induce discrete feature representations. Extensive experiments across various architectures and image datasets demonstrate that ICBM enables stable discretization while achieving prediction performance comparable to or better than standard models.

\emph{Scaling} has been the central driver of progress in modern AI, yet data curation remains an exception: it still relies on manual heuristics rather than systematic, scalable methods. In practice, data recipes (e.g., filter thresholds, deduplication strength, domain mixing ratios) are determined through many small-scale ablation training runs. This work demonstrates that the performance of the best-discovered recipe improves \emph{predictably}--as a \emph{power law}--in the number of such ablation runs. Theoretically, the power-law form arises naturally from regret bounds in zero-order optimization theory. Empirically, we validate it across diverse training stages, including pretraining, supervised fine-tuning, preference optimization, and reinforcement learning. Building on this finding, we derive a principled approach to \emph{data quality--quantity tradeoff} under a fixed compute budget, yielding a Pareto frontier that dominates standard heuristic pipelines (e.g., DCLM). Together, this work provides a general framework for scaling data-curation compute, establishing it as a new axis alongside the well-studied pretraining and test-time scaling.

Personalization histories vary across domains because surface actions and objectives differ: users watch or skip movies, click or ignore news, or request recommendation, summarization, question answering, or response generation. Rather than build a separate history encoder for each action space, we rewrite native interactions into a shared *action-on-item event schema*: a specific action on a specific item becomes positive, negative, or command-like evidence over a target-locus embedding whose geometry induces item-kind abstraction. Task requests become generalized commands over the active item-side abstraction. The modeling problem is how this evidence enters a user-specific state, persists, and is read out by the active command. We formulate the Multi-Timescale State Hypothesis (MTSH), where signed and command-like evidence flows through long-term stable interests, recency-sensitive short-term interests, and bursty episodic traces. We instantiate MTSH with $\texttt{PerTIDE}$, an action-conditioned multi-timescale encoder that learns a user-specific preference-history state and uses command-conditioned readout to produce a task-ready state. We evaluate $\texttt{PerTIDE}$ on MovieLens, MIND, and PENS for prediction, and PENS/OpenAI-Reddit for generation. $\texttt{PerTIDE}$ improves over the strongest baselines without explicit trace factorization by +2.36/+3.47, +10.05/+9.01, and +0.90/+1.42 MRR/nDCG on MovieLens, MIND, and PENS; and over the strongest two-shot large language model by +0.10/+0.18 and +0.04/+0.19 PerSEval-JSD/PerSEval-METEOR on PENS and OpenAI-Reddit. Removing action conditioning and command-conditioned readout degrades MIND by 6.52 MRR / 9.13 HR@10 and 4.14 MRR / 4.42 HR@10. Trace diagnostics show L/S/E specialization, while the full $L+S+E$ model is strongest under mixed stable, recency-sensitive, and episodic command evidence. These results support action-conditioned multi-timescale preference flow as a reusable user-history encoding principle.

Standard flow and diffusion pre-training matches the distribution of available data (e.g., molecules), which often covers only a small fraction of the valid design space. In generative discovery, however, one aims to sample valid new-to-nature designs, assigned negligible probability under, and thus inaccessible to, standard models fitted to the observed data. To overcome this limitation, we depart from data distribution matching and view a generative model through its generable set: the region it covers with non-negligible probability. This allows to introduce a new learning principle for out-of-distribution flow modeling: enlarging a model’s generable set to increase coverage of the valid design space. We propose Active Flow Expansion (ActFlow), a continued pre-training method that employs verifier feedback to expand a pre-trained model over new valid regions by iteratively adapting to synthetic data generated through active exploration in the learned flow representation. Theoretically, we establish to our knowledge first-of-their-kind statistical learning guarantees for out-of-distribution flow modeling, analyzing generable set expansion as a local-to-global reachability process over a learned representation. Empirically, we assess ActFlow with suitable out-of-distribution generative modeling metrics across small organic molecules, mid-sized drug-like molecules, therapeutic peptides, and protein sequence design tasks. Results show that ActFlow expands valid coverage far beyond the region modeled by the initial pre-trained model, significantly outperforming widely adopted synthetic flow pre-training methods.

While recent multimodal recommender systems have demonstrated the effectiveness of incorporating visual and textual information to improve downstream performance, most existing methods rely on static modality fusion, assuming that the relative importance of textual and visual signals remains stable across recommendation scenarios. This design may not fully account for an important variation across recommendation requests: some queries require fine-grained visual cues, whereas others are better served by textual or functional semantics, in which case indiscriminate modality fusion brings in uninformative cues and impairs recommendation quality. To address this, we propose AdaM-Rec, an LLM-based framework for adaptive modality routing in multimodal recommendation, which enables dynamic calibration of reliance on textual and multimodal evidence for user-specific queries. Built on structured natural-language representations of items and user preferences, it estimates modality reliability using proxy recall tasks. Specifically, it generates pseudo-queries that match the granularity of the actual query while pointing to the user's positively interacted items as verifiable proxy targets, evaluating which modality yields better recall performance in analogous scenarios and optimizing the routing strategy in an agentic manner. It then performs routed recall with optimized strategy, enriches results with collaborative items, and ranks candidates by their relevance to both the query and user preferences. Experiments demonstrate that AdaM-Rec delivers strong performance against state-of-the-art baselines, highlighting the effectiveness and broader potential of adaptive control over modality reliance in multimodal recommendation. Code will be publicly available.

A major bottleneck in characterizing the failure modes of generative AI systems is the cost and time of annotation and evaluation. Consequently, adaptive testing paradigms have gained popularity, where one opportunistically decides which cases and how many to annotate based on past results. While this framework is highly practical, its extreme flexibility makes it difficult to draw statistically rigorous conclusions, as it violates classical assumptions: the number of observations is typically limited (often 10 to 50 cases) and decisions regarding sampling and stopping are made in the midst of data collection rather than based a pre-specified rule. To characterize what statistical inferences can be drawn from highly adaptive audits, we introduce a hypothesis testing framework from two 'dueling' perspectives: (i) the model's null that asserts there is no failure mode with performance below a target threshold versus (ii) the auditor's null that asserts they have a sampling strategy that will uncover a failure mode. Leveraging Safe Anytime-Valid Inference (SAVI), we formalize the auditor as conducting 'testing by betting', which translates into simultaneous e-processes for testing the dueling null hypotheses. Furthermore, if the auditor is sufficiently powerful, we prove that these two hypotheses are asymptotically inverses of each other, in that passage of a stringent audit does in fact certify the AI system as being globally robust. Empirically, we demonstrate that our proposed testing procedures maintain anytime-valid type-I error control, outperform pre-specified testing methods, and can reach statistically rigorous conclusions sometimes with as few as 20 observations.

Large Reasoning Models (LRMs) produce explicit chains of thought to improve transparency, but they frequently overthink, generating long yet inefficient reasoning chains that inflate computational costs and risk introducing errors. While numerous reinforcement learning methods struggle to penalize verbosity to achieve conciseness, they typically apply the penalty uniformly across lengthy sequences, which leads to \textit{credit confounding}, that is, the failure to distinguish essential reasoning steps from superfluous ones. In this paper, we study a measurable proxy for this confounding: high-entropy tokens that often coincide with surface transition points where reasoning re-enters verification. Based on this insight, we propose Adaptive Entropy-Sparing (AES), which selectively spares high-entropy tokens that occur before an efficient in-group reference, while penalizing overlength high-entropy suffix tokens associated with superfluous branching. Our AES thereby navigates the suppression-exploration trade-off, reducing overthinking while preserving average reasoning accuracy. We also present a stylized entropy-length model that motivates why local uncertainty can serve as a proxy for future reasoning cost. Across eight reasoning benchmarks, AES reduces reasoning length by up to 52.9\% while increasing mean accuracy by up to 3.7\% over the base models, consistently outperforming existing efficiency-oriented methods.

Temporal action segmentation (TAS) is commonly trained on large banks of pre-extracted frame features rather than raw RGB video, but storing and replaying those features becomes a practical bottleneck for long-form video, especially in incremental TAS where past-task data must remain available to mitigate catastrophic forgetting. Recent TAS replay and condensation methods reduce this burden, but they rely on dense frame-wise annotations and allocate storage through fixed segment-level structures, limiting their flexibility for long procedural videos whose frame complexity varies substantially over time. We introduce Adaptive Residual-Quantized Variational AutoEncoder (ARQ-VAE), a compact and label-free memory representation for TAS features that learns a discrete residual quantization of frame-level video features and then adaptively truncates the residual chain on a per-frame basis, retaining only the residual prefix that best reconstructs each frame feature. This yields a practical compact feature memory that does not depend on frame-wise labels and can therefore support multiple TAS settings with the same stored representation, while a lightweight fine-tuning stage further improves reconstruction quality for the truncated codes that are actually retained. Across standard TAS benchmarks, ARQ-VAE achieves a substantially improved storage--performance trade-off over prior replay and condensation baselines. On Breakfast, our final representation reduces training-set storage from 28 GB for the original I3D feature bank to 10 MB while retaining strong downstream segmentation performance; with ASFormer, it achieves 71.5 Edit and 51.2 F1@50 in the supervised setting. We further show that the same label-free memory remains effective for incremental and unsupervised TAS, and transfers to an alternative pretrained feature space based on DINOv3. We position ARQ-VAE as a practical memory representation for TAS, and a promising direction for broader dataset condensation applications.

Recommendation unlearning (RU) aims to remove the influence of specified user--item interactions from a trained recommender while preserving utility on the remaining data. Existing RU methods usually require access to the original training data, especially the remaining data, to recover recommendation utility after unlearning. However, access to such data is often restricted or unavailable in practice due to privacy constraints, storage costs, or third-party deployment, giving rise to source-free recommendation unlearning. To address this challenge, we propose an adaptive source-free recommendation unlearning (AdaSRU) method that enables effective unlearning without accessing the original remaining data. AdaSRU first estimates the inaccessible gradient on the remaining data. It exploits the first-order optimality condition of the trained recommender and approximates the remaining gradient using the forgetting data and squared gradient accumulators from adaptive optimizers, e.g., Adam. AdaSRU then formulates unlearning as a gradient-constrained update, where the estimated remaining gradient is used as a constraint to limit utility degradation while maximizing the unlearning objective. AdaSRU adaptively adjusts the utility-preserving correction according to the current gradient geometry. We further theoretically prove that AdaSRU approaches an approximate Pareto stationary solution. Extensive experiments on three real-world datasets demonstrate that AdaSRU achieves strong performance in both unlearning and recommendation. The code is available at https://anonymous.4open.science/r/AdaSRU-94D9/.

Human players in cooperative games often rely on conventions: reusable patterns of play that coordinate well when used by the whole team, but can fail otherwise. Evaluating agents in this setting is difficult because human demonstration datasets are usually unlabeled: trajectories do not identify which player, group, or convention produced them. As a result, standard proxy-based evaluations can miss important population diversity. In particular, proxy bots produced by regularised self-play around a pooled behavioural cloning policy can remain concentrated around a single dominant mode of play. We introduce a pipeline for recovering and evaluating this hidden convention diversity from unlabeled demonstrations. Our method first learns latent convention structure from the dataset, then selects a compact set of distinct but plausible conventions, and finally reconstructs each selected convention into a high-performing playable proxy using KL-regularised PPO. A key practical challenge is that KL-regularised PPO is sensitive to the KL penalty, which is often tuned using online interaction. We provide an empirical offline calibration procedure for choosing this KL penalty in the PPO+KL reconstruction settings studied here. On Hanabi, we find that the BC-derived proxy suite is highly concentrated, and that the relative ordering of agents changes when evaluation uses the latent proxy suite instead.

Neural contextual bandits support high-stakes decision systems---from clinical trial allocation to content recommendation. This paper investigates an adversarial threat, group-targeted suppression attacks (GTSA), in which an adversary selectively degrades performance for a demographic subgroup while preserving global performance metrics. We first show that under standard aggregate monitoring, GTSA with budget $B = \Omega(\sqrt{T})$ can induce $\Omega(T)$ group regret while maintaining only $o(1)$ deviation in observable global statistics, establishing that any defense that ignores group structure is fundamentally vulnerable. To address this vulnerability, we propose AnchorFair, a robust mitigation framework comprising three components: (i) a trust-weighted policy anchor with provable bounded drift, (ii) representation-stable demographic discovery via online clustering in a slowly evolving embedding space, and (iii) a demographic-aware exploration mechanism that adaptively amplifies learning for underrepresented or low-trust groups. We prove that \textsc{AnchorFair} achieves sublinear group regret $\tilde{O}(\sqrt{T})$ for all groups under GTSA.

We propose FAR-SIGN (Fully Asynchronous Robust optimization via SIGNed directional projections) for adversary-resilient learning in parameter-server--worker systems. FAR-SIGN achieves robustness through sign-based updates along carefully designed directions and mitigates the resulting bias via a two-timescale mechanism. It admits both first-order and zeroth-order implementations and enables fully asynchronous execution without requiring a private reference dataset at the server. We establish almost-sure convergence of FAR-SIGN to the set of stationary points for smooth, nonconvex objectives. Moreover, we prove the near-optimal rate of $O(n^{-1/4+\epsilon})$ in the first-order setting and the standard $O(n^{-1/6+\epsilon})$ in the zeroth-order setting, where $n$ is the iteration count and $\epsilon>0$ can be chosen arbitrarily small. Experiments on MNIST show that FAR-SIGN outperforms robust aggregation-based methods in both accuracy and wall-clock time.

LLM-as-a-Judge systems supply the reward signal in modern RLHF and RLVR pipelines, but their binary verdict reduces to a single linear readout on one hidden state. We show this readout is shallow enough that short, low-perplexity tokens flip the verdict from "No" to "Yes". These tokens are sampled from the judge's own next-token distribution at the response position, with no manual seed set and no gradient-based optimization. Our procedure, AdvJudge-Zero, reaches $>$90\% ensemble false-positive rate on 22 of 24 (model, dataset) cells across six Qwen, Llama, and Gemma judges, versus 54-72\% for the prior curated 10-token benchmark, and the discovered surface transfers cross-format to a 70B scalar reward model. The same discovered pool enables a defense: a LoRA fine-tune stratified by a 9-class mechanism taxonomy hardens cross-family generalization where naive sampling on the same pool fails, with mechanism breadth rather than pool size carrying the gain. Under GRPO training, the hardened judge eliminates the reward-collapse failures (false-positive spikes and length collapse) we observe in the unhardened baseline on both MATH and GSM8K at ten seeds per condition. The discovered pool, the mechanism taxonomy, and per-prompt flip records will be released under responsible disclosure.

Atomic electron tomography (AET) aims to recover the three-dimensional arrangement, chemical identity, and temporal evolution of individual atoms. Yet most reconstruction methods are still evaluated primarily by voxel-level image similarity, such as PSNR, SSIM, or Fourier shell correlation. In this work, we show that this evaluation practice can select scientifically incorrect reconstructions: a volume that looks accurate can contain atoms that are shifted, merged, missed, hallucinated, chemically mislabeled, or assigned inconsistent identities through time. We introduce AET-Bench, a dataset and benchmark for atom-first evaluation of atomic electron tomography. AET-Bench contains 755 samples across a six-tier realism ladder, spanning clean analytical simulations, binary alloys, 4D atomic dynamics, multislice scattering, real experimental tilt series, and Brownian Pt nanocrystal trajectories. It defines four complementary tasks and evaluates 14 core method families with 21 logged method entries under a unified protocol. AET-Bench reveals a central failure mode hidden by voxel-only evaluation. On the AET-Sim-Mono tier, high-SSIM neural and Gaussian reconstructions can recover substantially fewer atoms than classical or sparsity-regularized methods: for example, 4D-GS achieves strong voxel similarity but only 43.9\% atom recall, while FISTA-TV reaches 95.1\% atom recall despite much lower SSIM. This inversion persists across atom-matching thresholds, showing that it is not an artifact of a single cutoff. On more realistic multislice and real tiers, atom recovery further degrades, exposing a physics gap between visually plausible volumes and reference atomic models. For 4D reconstruction, we find that smooth temporal density fields do not guarantee persistent atom identities. AET-Bench establishes atom-level fidelity as a necessary evaluation axis for AET. Our results suggest that future AET methods should not be judged only by how reconstructions look, but by whether they recover the atoms, species, and trajectories that scientists actually use.

Few-step diffusion sampling sharpens the tension between ODE and SDE dynamics. ODE samplers are stable but can lose diversity, while the standard SDE endpoint can become unreliable at low numbers of function evaluations (NFE). The challenge is therefore not simply whether stochasticity should be added, but which driver properties allow stochastic correction to contribute diffusion without losing finite-step control. We show that this tradeoff is governed by two driver-level quantities, the asymptotic covariance $Q$, which sets the effective diffusion level on the ODE-to-SDE spectrum, and the Poisson solution norm $C_\chi$, which controls the finite-step remainder. Gaussian drivers have $C_\chi = \infty$, so this finite-step remainder is not uniformly controlled by the same bound. Balancing the diffusion level set by $Q$ with the finite-step control set by $C_\chi$ formalizes a favorable regime for bounded structured drivers with intermediate diffusion strength. This regime directly gives a structured correction method, instantiated as Fast-Driver Sampler (FDS) with the shuffled-Lorenz driver. Across DiT-XL/2 and EDM2, FDS establishes a stronger low-NFE frontier among evaluated training-free samplers, improving ODE baselines while avoiding the degradation of the SDE endpoint.

The rapid scale-up of AI infrastructure has made datacenter reliability a cornerstone of the cloud business economy. Hardware failures in hyper-scale AI clusters do not merely disrupt individual machines; they stall multi-million dollar training workloads, and so the downtime of individual machines can cause outsized economic losses. Minimizing machine downtime, therefore, is more important than ever. To address this, we present DrWatson, an AI-driven hardware diagnosis system that pioneers a foundation-model approach to automated datacenter maintenance. Deployed at production scale within a major AI hyperscaler, DrWatson yields substantial efficiency gains across diverse compute architectures. Real-world evaluations demonstrate that it reduces production downtime by 18.3% on GPU platforms and 16.1% on custom AI accelerators. Furthermore, it significantly accelerates hardware deployment, cutting quality assurance (QA) time by 28.3% for GPUs and 25.1% for custom accelerators. These results establish DrWatson as a field-tested solution for maximizing the reliability and cost efficiency of next generation AI fleets.

Contextual integrity (CI) holds that appropriate information flow depends on context-specific norms. The same disclosure may be appropriate in one context and inappropriate in another. As actors handle private user data in multi-actor pipelines, we ask whether they track these norms faithfully under social and contextual pressure. We build a generative model of CI in multi-actor systems: scenarios drawn from CI's taxonomies, a generation pipeline, multi-actor simulation, and a two-sided appropriateness metric. The metric scores both withholding when norms forbid sharing and sharing when they permit it. Across three intervention axes (decision logic, cognitive profile, and model choice), no actor-level lever closes both sides of the appropriateness gap; protection and utility trade off in every variant we tested. Closing the gap may require mechanisms beyond the actor level, such as CI-grounded policy and institution design.

Symbolic world models (e.g., PDDL domains or executable simulators) are central to model-based planning, but training LLMs to generate such world models is limited by the lack of large-scale verifiable supervision. Current approaches rely primarily on static validation methods that fail to catch behavior-level errors arising from interactive execution. In this paper, we propose Agent2World, a tool-augmented multi-agent framework that achieves strong inference-time world-model generation and also serves as a data engine for supervised fine-tuning, by grounding generation in multi-agent feedback. Agent2World follows a three-stage pipeline: (i) A Deep Researcher agent performs knowledge synthesis by web searching to address specification gaps; (ii) A Model Developer agent implements executable world models; And (iii) a specialized Testing Team conducts adaptive unit testing and simulation-based validation. Agent2World demonstrates superior inference-time performance across three benchmarks spanning both Planning Domain Definition Language (PDDL) and executable code representations, achieving consistent state-of-the-art results. Beyond inference, Testing Team serves as an interactive environment for the Model Developer, providing behavior-aware adaptive feedback that yields multi-turn training trajectories. The model fine-tuned on these trajectories substantially improves world-model generation, yielding an average relative gain of 30.95% over the same model before training.

The rise of generative AI and autonomous agents is creating ecosystems in which stakeholders strategically choose which agents to deploy on their behalf. These design choices--such as model backbone, prompting strategy, tool access, and fine-tuning pipeline--are often strategically interdependent, as the performance of a deployed agent depends on the agents chosen by others. This position paper argues that this agent design stage should itself be a target for mediation. This perspective shifts attention from mediating only the \emph{downstream} interaction among deployed agents to mediating the \emph{upstream} stakeholder game that determines which agents enter that interaction in the first place. We formalize this agent design stage as a \emph{meta-game}, show how its equilibria can be collectively harmful, and discuss how various forms of mediation can realign incentives toward socially beneficial outcomes. We demonstrate these mediation approaches using both stylized examples and a real-world case study constructed from empirical LLM-agent interaction data. We outline concrete research directions for developing mediators that steer agentic design choices toward socially desirable outcomes.

Formal Geometric Problem Solving requires every step of reasoning within a strict logical framework, and has been a core challenge in artificial intelligence. This paper presents a unified neuro-symbolic reasoning framework, named Reflect Agent Verified Solve (RAVS), which deeply integrates large language models, agent architectures, and formal symbolic solvers. The core innovations of RAVS lie in two parts. First, we propose a role-separated neuro-symbolic collaboration mechanism: the LLM serves as a planner responsible for high-level semantic understanding and path reflection, while the symbolic solver acts as an executor responsible for formal verification and rigorous theorem application. The neural reasoning capability of the LLM and the logical completeness of the symbolic system complement each other, fundamentally eliminating the risk of hallucinations. Second, we construct the first bidirectional symbolic reasoning engine that fully unifies forward and backward solving. For each theorem, we define two reversible operations: forward application and backward decomposition. We further provide a theoretical analysis that bidirectional reasoning achieves an exponential complexity advantage over unidirectional reasoning. Together, these two mechanisms form an iterative closed loop between neural planning and symbolic execution. On the FormalGeo7K benchmark, RAVS achieves a 95.8% solving accuracy, substantially outperforming existing state-of-the-art methods, without requiring any additional problem-specific annotated data.

Agentic data science (ADS) systems are rapidly improving their capability to autonomously analyze, fit, and interpret data, potentially moving towards a future where agents conduct the vast majority of data-science work. However, current ADS systems use statistical tools designed to be interpretable by humans, rather than interpretable by agents. To address this, we introduce Agentic-imodels, an agentic autoresearch loop that evolves data-science tools designed to be interpretable by agents. Specifically, it develops a library of scikit-learn-compatible regressors for tabular data that are optimized for both predictive performance and a novel LLM-based interpretability metric. The metric measures a suite of LLM-graded tests that probe whether a model’s string representation is “simulatable” by an LLM, i.e. whether the LLM can answer questions about the model’s behavior by reading its string output alone. We find that the evolved models jointly improve predictive performance and interpretability, generalizing to new datasets and new interpretability tests. Furthermore, these evolved models improve downstream end-to-end ADS, increasing performance for Copilot CLI, Claude Code, and Codex on the BLADE benchmark by up to 73%.

We take the position that agent security must be approached as a systems problem: the AI model powering the agent must be treated as an untrusted component, and security invariants must be enforced at the system level. Through this lens, efforts to increase model robustness (the dominant viewpoint in the community) are insufficient on their own. Instead, we must complement existing efforts with techniques from the systems security domain. Based on our experience as cybersecurity researchers in operating systems, networks, formal methods, and adversarial machine learning, we articulate a set of core principles, grounded in decades of systems security research, that provide a foundation for designing agentic systems with predictable guarantees. As evidence, we analyze eleven representative real-world attacks on agents and discuss how systems principles, if realized, could have prevented these attacks. We also identify the research challenges that stand in the way of implementing these principles in agents.

Real-world multimodal agents solve multi-step workflows grounded in visual evidence. For example, an agent can troubleshoot a device by linking a wiring photo to a schematic and validating the fix with online documentation, or plan a trip by interpreting a transit map and checking schedules under routing constraints. However, existing multimodal benchmarks mainly evaluate single-turn visual reasoning or specific tool skills, and they do not fully capture the realism, visual subtlety, and long-horizon tool use that practical agents require. We introduce AgentVista, a benchmark for generalist multimodal agents that spans 25 sub-domains across 7 categories, pairing realistic and detail-rich visual scenarios with natural hybrid tool use. Tasks require long-horizon tool interactions across modalities, including web search, image search, page navigation, and code-based operations for both image processing and general programming. Comprehensive evaluation of state-of-the-art models exposes significant gaps in their ability to carry out long-horizon multimodal tool use. Even the best model in our evaluation, GPT-5.4 with tools, achieves only 31.58% overall accuracy, and hard instances can extend to 25 tool-calling turns. We expect AgentVista to accelerate the development of more capable and reliable multimodal agents for realistic and ultra-challenging problem solving.

Message-passing neural networks (MPNNs) learn node representations by iteratively aggregating information from neighbors, but stacking too many layers causes all representations to converge, a phenomenon known as oversmoothing. We propose a geometric perspective on this phenomenon. At each depth $k$, the propagation operator assigns to every node $v$ a walk distribution $p_v^{(k)}$, a probability distribution over all nodes describing how $v$ distributes its attention across the graph. These walk distributions form a point cloud on the probability simplex, and oversmoothing is the contraction of this cloud to a single point. To measure this contraction, we exploit the intrinsic geometry of the simplex: the Fisher--Rao metric, the unique Riemannian metric invariant under sufficient statistics, whose chordal distance under the square-root embedding $p \mapsto \sqrt{p}$ is the Hellinger distance. The resulting diagnostic, the \emph{aggregation dispersion} $\mathcal{D}_k$, is the variance of the embedded point cloud on the unit sphere. We prove that: (i) $\mathcal{D}_k = 0$ if and only if the $k$-th step propagation matrix has rank one, providing a necessary and sufficient condition for representational collapse; (ii) $\mathcal{D}_k$ is monotonically non-increasing with depth, so each layer irreversibly spends a finite diversity budget; (iii) the spectral gap of the propagation matrix controls the rate of this decay; and (iv) for contractive MPNNs, the aggregation dispersion upper-bounds the representation dispersion up to architecture-dependent constants, connecting the geometry of the simplex to the geometry of the feature space. In experiments across 14 model configurations and 11 datasets, $\mathcal{D}_k$ achieves the highest average Pearson correlation with accuracy degradation among all training-free diagnostics ($|r| = 0.721$), outperforming the post-training effective rank ($|r| = 0.688$).

Recent works on language identification and generation have established tight statistical rates at which these tasks can be achieved. These works typically operate under a strong realizability assumption: that the input data is drawn from an unknown distribution necessarily supported on some language in a given collection. In this work, we relax this assumption of realizability entirely, and impose no restrictions on the distribution of the input data. We propose objectives to study both language identification and generation in this more general ''agnostic'' setup. Across both problems, we obtain novel interesting characterizations and nearly tight rates.

Graph clustering is a fundamental task for understanding graph-structured data without labels, yet classical end-to-end methods require retraining on each dataset, limiting their generalization ability. Although Graph Foundation Models (GFMs) enable transferable learning across diverse graph tasks, they are not directly applicable to graph clustering. This limitation stems from two key factors: the inability to learn mutually beneficial knowledge on both the node and graph levels, and the lack of a completely unsupervised clustering-oriented design. To bridge this gap, we propose a Graph Foundation Model for Unified Clustering (GFM-UC), which pre-trains on the source domain and only requires fine-tuning for fast clustering in the target domain. The entire learning process does not require any labels. Specifically, in the pre-training stage, we select learnable high-confidence unified subgraph prototypes from the source domain. And then we generalize them to train the clustering network collaboratively. In the fine-tuning stage, we align the target domain with the source domain in a distribution-aware manner to further refine the clustering head and obtain more well-separated cluster boundaries. Extensive experiments on five datasets demonstrate that our method significantly outperforms existing state-of-the-art approaches.

Learning high-quality multimodal entity representations is essential for advancing reasoning tasks such as multimodal knowledge graph completion (MMKGC). However, rigid alignment in existing fusion strategies can bias representations toward a dominant anchor modality, weakening fine-grained complementary cues and propagating noisy signals into relational reasoning. To address these limitations, we propose MANoR, a Multimodal Autonomous Negotiation framework for calibrated Representation learning. MANoR treats modalities as autonomous semantic agents that negotiate before fusion. Specifically, MANoR preserves modality-specific semantics through centroid-guided subspace contraction, which reduces unstable centroid-relative dispersion while retaining entity-level residual semantics. On this stabilized basis, negotiated attention enables selective cross-modal interaction without collapsing modality boundaries. As modalities can still differ in reliability after interaction, uncertainty calibrated fusion estimates dimension-level aleatoric uncertainty and modality-level reliability to compute relation-aware fusion weights. In this way, MANoR couples modality autonomy, selective interaction, and calibrated integration within a unified representation learning framework. Experiments on MKG-W, MKG-Y, TIVA, and KVC16K show that MANoR consistently outperforms strong MMKGC baselines, achieving a relative Hits@1 gain of up to 17.72% on KVC16K.

Agricultural management is heterogeneous along two axes. The first is distributional shift within a fixed task interface: the same sensors, action menu, and objective, but with weather, planted crops, and prices varying across years and locations. The second is structural shift across interfaces: different farms expose different observed variables, available actions, and management objectives. We formalize the second axis in MDP terms as \emph{schema shift}: variation in the policy-facing schema $\Sigma = (\Sigma_O, \Sigma_A, \Sigma_R)$ between training and deployment. Fixed-interface neural policies are structurally inapplicable when $\Sigma_O$ or $\Sigma_A$ shifts and require retraining when $\Sigma_R$ shifts, while prompt-conditioned LLM policies re-read $\Sigma$ from the prompt of each new episode. Whether LLM-RL actually generalizes across both axes has not been systematically tested. We address this question with \emph{AgriManager}, the first multi-simulator LLM-RL framework for agricultural management, unifying the DSSAT, WOFOST, and Cycles crop simulators under prompt-based schema specification with joint multi-objective, multi-environment training. On this platform we design a three-tier generalization ladder, each tier grounded in a real-world deployment scenario: Tier 1 within-schema distributional shift (weather, crop, price), Tier 2 single-axis schema shift along $\Sigma_O / \Sigma_A / \Sigma_R$, Tier 3 compositional schema shift across simulators. Within fixed schemas, prompt-conditioned LLM-RL with explicit reasoning matches or exceeds fixed-interface NN-RL on all three settings, with reasoning becoming necessary when the reward shape penalizes input use. Single-axis schema shift imposes a different burden along each axis: information sufficiency for $\Sigma_O$, joint action-menu coverage for $\Sigma_A$, and objective coverage for $\Sigma_R$. Under compositional schema shift, a unified policy with marginal per-platform coverage and explicit reasoning matches or exceeds per-source specialists on all three callback targets. Without that coverage, pairwise zero-shot transfer is bounded by transition dynamics, which the prompt cannot describe and which we leave as an open task. Together, these results provide an initial empirical baseline for unified prompt-conditioned control across heterogeneous agricultural deployments.

There are now thousands of measurement instruments used to evaluate language models, and the resulting scores are routinely used to support consequential claims about capabilities and risks. Yet, the evidence linking benchmark scores to constructs (i.e., abstract concepts such as reasoning ability or safety) is often implicit, incomplete, or absent. We introduce AI Construct Lexis, a versioned ontology and interactive artifact that extracts and organizes relationships among constructs, behaviors, and measurement instruments as claimed in the literature for LLM evaluation. Drawing on over 8,337 published papers from selected top AI venues, we use language models to extract constructs, definitions, behavioral indicators, and measurement instruments, yielding 977 distinct constructs, 1,694 measurement instruments, 830 behavioral indicators, and 5,900 relationships. The Lexis captures relationships claimed in the published literature, not necessarily validated relationships; valid score interpretation still requires empirical evidence that the instruments capture the constructs they are used to support. By exposing the field’s implicit measurement assumptions, the Lexis enables their inspection, testing, falsification, and refinement into evidence-based relationships. On expert-annotated samples, the extraction pipeline achieves high agreement on extracted concepts while revealing substantial ambiguity in the boundaries of constructs and behaviors. By making these relationships explicit, AI Construct Lexis provides infrastructure for a more rigorous and cumulative science of AI evaluation.

We argue that in multiple areas of AI for drug discovery (AIDD), deep learning models are not learning the meaningful biological or chemical features they were hypothesised to capture, but instead learn non-generalisable features, for example, from dataset biases. To address this, we propose the systematic use of misaligned baselines. We define misaligned baselines as models that rely on signals purposely misaligned with the intended learning objective. Rather than modelling the underlying biology or chemistry directly, these baselines draw on reductionist sources of information, heavily perturbed input features, or heuristic models. Competitive performance of such baselines reveals when models are not learning the biologically meaningful representations they were designed to learn. By examining case studies across multiple branches of AIDD, we demonstrate that misaligned baselines have consistently exposed such failure modes and crucially informed the evaluation and improvement of these models. We argue that adopting them as standard practice will help to ensure that progress in AIDD reflects genuine advances rather than artefacts of data-generating processes or evaluation design.

Capability evaluations aim to surface dangerous model behaviors before deployment, but their reliability depends on the assumption that hidden capabilities can be elicited. We show this assumption does not hold under adversarial conditions by constructing backdoor attacks that embed encrypted neural networks within host model weights, decrypting and executing them only upon receiving a secret trigger. Using digital lockers, these hidden circuits remain provably hard to elicit and interpret under standard cryptographic hardness assumptions, even given full access to model weights, extending prior work from hiding fixed strings to arbitrary computations. We implement our constructions in PyTorch and empirically validate resistance to supervised fine-tuning on a small chemical reaction prediction task - a proxy for CBRN-relevant capabilities. Our results reveal a limit of capability audits. We release our models to support the development of stronger defenses.

We consider linear equations with min and max operators~(LEMMs) that contain many subproblems ranging from optimization, learning, to games. Recently, Chatterjee et al. [2025] gave a systematic study of the complexity of different subclasses. Three key subclasses--(I) halting branching process, (II) absolutely halting LEMMs, and (III) halting LEMMs--are proved to be in UP $\cap$ coUP while generalizing stochastic games (SSGs). In this work, we study the classic algorithms of Policy Iteration (PI) and Value Iteration (VI) for these general subclasses. First, we simplify the problem hierarchy by showing the equivalence between halting branching process and SSGs. Then, we show that while PI diverges for absolutely halting LEMMs due to the loss of monotonicity, VI remains convergent, a result we establish via diagonal rescaling. Finally, we show that neither PI nor VI converges for general halting LEMMs and, to this end, propose variants of simple policy iteration that ensure convergence across all subclasses.

Engineering LLMs to accelerate life sciences research requires a robust alignment with biomedical knowledge. We observe that biomedical text exhibits a fundamentally different uncertainty structure from general text: dense low-confidence runs encode epistemic knowledge gaps (dense causal chains, rare entities) rather than the sparse aleatoric stylistic variation typical of general text. Based on this discovery, we propose Balanced Fine-Tuning (BFT), a dual-scale post-training method that combines group-normalized token reweighting with sequence-level reallocation toward knowledge-dense samples exhibiting dense epistemic uncertainty. Across medical evaluation, biological reasoning, sparse-reward RL, and biological representation tasks, BFT provides more consistent gains than SFT and DFT under a shared training setup. When replacing the default closed-source backbones in GeneAgent (GPT-4o) and VCWorld (Gemini-2.5-Flash), the BFT-aligned 70B model delivers stronger performance across biological process reasoning and chemical perturbation prediction. Critically, all BFT variants further improve after subsequent GRPO with sparse rewards, while SFT and DFT degrade, suggesting that epistemic-aware post-training provides a more robust policy initialization. Beyond text generation, BFT-aligned LLMs produce more accurate and professional biomedical profile texts; after encoding these profiles with a text embedding model, the resulting representations support gene-level, cell-level, and perturbation-response tasks, suggesting that BFT-enhanced generation can facilitate biological representation and, in turn, broader biomedical downstream tasks.

Multimodal Large Language Models (MLLMs) have achieved impressive performance across a wide range of reasoning tasks, yet it remains unclear whether their internal representations and decision criteria align with the latent structure of human cognition. In this work, we propose a novel framework to align MLLMs with the multidimensional mental representations underlying human similarity judgements. We build on a previously established behavior-derived semantic embedding space, in which objects are represented along 66 interpretable dimensions derived from 4.70 million human Odd-One-Out (O1O) judgements. For each triplet, we infer the most salient decision dimension and convert it into a dimension-grounded linguistic rationale. This enables MLLMs to learn both the final choice and a justification consistent with the inferred behavioral criterion. Experimental results demonstrate that our approach improves model consistency with human similarity judgements while retaining competitive performance on standard multimodal benchmarks. Furthermore, using searchlight Representational Similarity Analysis (RSA) on an independent fMRI Natural Scenes Dataset (NSD), we observe increased representational alignment between model and human. Ultimately, our findings point to a data-driven route for incorporating human cognitive structure into MLLMs alignment.

Although Large Language Models (LLMs) achieve strong alignment through supervised fine-tuning and reinforcement learning from human feedback, the alignment is often fragile under subsequent fine-tuning. Existing explanations either attribute alignment fragility to gradient geometry or characterize it as a distributional shift in model outputs, yet few provide a unified account that bridges parameter-space learning dynamics with function-space alignment behavior during fine-tuning. In this work, we introduce a tractable alignment score and derive its closed-form update during fine-tuning, yielding a unified framework for alignment dynamics. Our analysis decomposes alignment updates into two competing components: a Rebound Force, governed jointly by the current alignment state and the narrowness of model distribution, and a Driving Force, determined by how the training distribution aligns with outcome-conditioned posteriors over aligned and non-aligned completions. This decomposition explains why prior alignment can be reversed by later fine-tuning and why narrower posterior structure strengthens such reversal. Moreover, our framework predicts a Rehearsal Priming Effect: prior alignment leaves a latent posterior imprint that amplifies the effective Driving Force upon re-exposure, leading to faster re-alignment. We validate these predictions across safety alignment, emergent misalignment, and sentiment settings, demonstrating consistent alignment reversal and accelerated re-alignment under re-exposure. In addition, controlled experiments in safety alignment confirm the predicted dependence of rebound strength on posterior narrowness. Together, these results provide a unified dynamical perspective on how alignment is disrupted and reactivated during LLM fine-tuning.

With the rising need for spatially grounded tasks such as Vision-Language Navigation/Action, allocentric perception capabilities in Vision-Language Models (VLMs) are receiving growing focus. However, VLMs remain brittle on allocentric spatial queries that require explicit perspective shifts, where the answer depends on reasoning in a target-centric frame rather than the observed camera view. Thus, we introduce $\textbf{Allocentric Perceiver}$, a training-free strategy that recovers metric 3D states from one or more images with off-the-shelf geometric experts, and then instantiates a query-conditioned allocentric reference frame aligned with the instruction’s semantic intent. By deterministically transforming reconstructed geometry into the target frame and prompting the backbone VLM with structured, geometry-grounded representations, Allocentric Perceiver offloads mental rotation from implicit reasoning to explicit computation. We evaluate Allocentric Perceiver across multiple backbone families on spatial reasoning benchmarks, observing consistent and substantial gains ($\sim$\textbf{10\%}) on allocentric tasks while maintaining strong egocentric performance, and surpassing both spatial‑perception-finetuned models and state‑of‑the‑art open‑source and proprietary models.

Translating high-level, under-specified human commands into coherent video is fundamentally challenging. Current video generation models lack explicit reasoning capabilities and typically fail to understand ambiguous instructions, frequently resulting in physically inconsistent and causally disjointed generation. To address this, we introduce AmbiguousWorld, a novel inference-time reasoning framework that transforms video generation into a multi-modal Tree of Thoughts. We employ a Vision-Language Model as both a semantic planner to dynamically decompose fuzzy instructions into actionable subgoals, and as a closed-loop critic to prune invalid trajectories by evaluating speculative paths across multiple physical and functional dimensions. To systematically assess this, we annotate the first embodied video generation benchmark targeting ambiguous instructions based on the DROID dataset, and design a multi-dimensional VLM evaluation framework. Empirical results on state-of-the-art baselines demonstrate that our framework yields substantial improvements, boosting absolute Task Success by up to 29.0\% and the overall generative consistency weighted score by 24.2 points. Furthermore, we validate the effectiveness of our VLM evaluation framework through rigorous comparison with human assessment.

Recent foundation models (FMs) for zero-shot reconstruction of dynamical systems (DS) achieve strong out-of-domain generalization but provide little insight into the mechanisms that underlie their forecasts. Such an understanding could help to strip down overladen FM architectures to their bare essence and expose the minimal requirements for in-context learning in the DS domain. Toward this goal, here we iteratively reduce a recent powerful SOTA model for DS reconstruction, DynaMix (Hemmer & Durstewitz, 2025), to a minimal interpretable two-parameter form, which we call DynaBase. DynaBase produces forecasts through a linear blend of the current latent state and the nearest in-context neighbor and its temporal successor. Surprisingly, despite its extreme simplicity, DynaBase produces highly competitive zero-shot DS reconstructions across chaotic and cyclic systems, with a negligible parameter load, many orders of magnitude below that of other FMs. Even more, this extreme simplicity permits direct model optimization on DS reconstruction measures, as well as closed-form one-step analytical solutions on prediction MSE. Theoretical and empirical analysis of DynaBase further leads to a 1-parameter family of maps, with the context-parroting algorithm of (Zhang & Gilpin, 2026) recovered at one end, and chaotic (divergent but bounded) behavior at the other. We further show how different training strategies lead to models either optimal for short-term prediction or for DS reconstruction. Thus, DynaBase not only exposes the minimal mechanisms required for producing zero-shot DS reconstruction, but also reconciles within an accessible mathematical frame divergent observations in the literature.

We study image inpainting with generative diffusion models. Existing methods typically either train dedicated task-specific models, or adapt a pretrained diffusion model separately for each masked image at deployment. We introduce a middle-ground model, termed Amortized Inpainting with Diffusion (AID), which keeps a pretrained diffusion backbone fixed, trains a small reusable guidance module offline, and then reuses it across masked images without per-instance optimization. We formulate it as a deterministic guidance problem with a supervised terminal objective. To make this problem learnable in high dimensions, we derive an auxiliary Gaussian formulation and prove that solving this randomized problem recovers the optimal deterministic guidance field. This bridge yields a principled continuous-time actor--critic algorithm for learning the guidance module in a fully data-driven manner. Empirically, on AFHQv2 and FFHQ under the pixel EDM pipeline and on ImageNet under the latent EDM2 pipeline, AID consistently improves the quality--speed trade-off over strong fixed-backbone and amortized inpainting baselines across multiple mask types, while adding less than one percent trainable overhead.

Planner-guided model-based reinforcement learning (MBRL) has emerged as a powerful paradigm for continuous control, combining learned world models with online planning to achieve strong performance and sample efficiency. However, existing methods typically distill planner-improved actions into a unimodal Gaussian policy and reuse it as a proposal prior for subsequent planning, overlooking the fact that Model Predictive Path Integral (MPPI) planners can induce multimodal supervision over high-value actions. Compressing such supervision into a unimodal distribution can lead to mode averaging, limited proposal coverage, and unstable bootstrapped policy learning. To address this issue, we propose GeMP, a \textbf{Ge}nerative \textbf{M}ultimodal \textbf{P}lanning framework for MBRL. We theoretically characterize the multimodality of planner-induced supervision, showing that finite-sample planning can provide a distribution of high-value actions rather than a single deterministic target. To capture this distribution without compromising training stability, GeMP jointly learns a multimodal flow policy through value-weighted supervision and retains an auxiliary Gaussian policy for stablizing off-policy actor-critic learning. We further employ a heterogeneous proposal generator that combines Gaussian, flow-based, and random action-sequence candidates to improve proposal coverage for MPPI refinement. Experiments on MyoSuite and DeepMind Control Suite show that GeMP improves control performance and training stability over competitive baselines.

Echocardiography segmentation plays a central role in quantitative cardiovascular assessment. However, automated echocardiography segmentation remains challenging due to speckle noise, poorly defined anatomical boundaries, acquisition variability, and the limited anatomical coverage of existing benchmarks. Conventional segmentation methods demonstrate limited accuracy under such low image quality and domain shifts. Inspired by the observation that expert cardiologists can maintain reliable anatomical delineation by leveraging their structural understanding of the cardiac anatomy, we propose that incorporating anatomical priors into echocardiography segmentation models can improve both precision and robustness. To this end, we propose AST-Seg, an anatomy-aware spatio-temporal framework for echocardiography segmentation. AST-Seg introduces an anatomy-aware feature encoder that regularizes transformer attention with anatomy-derived spatial-relation priors, and a Deformable Spatio-Temporal Mamba module that captures localized and periodic cardiac motion through adaptive spatial aggregation followed by temporal state-space modeling. In addition, we curate an Echocardiography Anatomy (EA) dataset with pixel-level annotations for 12 clinically relevant cardiac anatomy, enabling comprehensive multi-structure cardiac interpretation. Extensive experiments on in-distribution CAMUS and EA datasets, as well as an out-of-distribution point-of-care echocardiography dataset, demonstrate that AST-Seg provides domain-generalized segmentation with improved precision.

Previous work has shown that the optimal classifier from a hypothesis class subject to exact fairness constraints may not be *robust*; that is, its accuracy can change drastically under small shifts to the underlying data distribution. We ask the following questions: Given a hypothesis class $\mathcal{H}$, is the accuracy of the optimal fair classifier from $\mathcal{H}$ robust to malicious distribution shifts? And is this property efficiently testable? We consider $\mathcal{H}$ that can be represented by a convex polytope — the natural setting for randomized ensembles of deterministic classifiers, including those produced by boosting. We provide a complete geometric characterization of when $\mathcal{H}$ is robust, and an efficient linear programming test to audit the robustness of a given $\mathcal{H}$. Surprisingly, we demonstrate that if we relax exact fairness constraints and only require approximate fairness, every $\mathcal{H}$ is robust. Our results hold for a broad class of linear fairness criteria (e.g., demographic parity, equal opportunity, predictive equality) and linear loss functions (e.g., 0/1 loss, weighted loss).

We propose a new framework for low-precision neural networks based on angular geometry and randomized low-bit estimators. Rather than approximating Euclidean linear operations under limited precision, we reinterpret neural computation in terms of directional similarity and construct quantized random-feature estimators of cosine interactions. Our approach introduces \textit{Angular Layers}, which replace standard linear transformations with low-bit projections that estimate angular similarity in the forward pass, while gradients are computed with respect to the underlying continuous cosine geometry. This decoupling enables stable optimization without straight-through estimators or heuristic surrogate gradients, even under aggressive quantization of both weights and activations. On the theory side, we introduce \emph{G-Gradient} as a gradient proxy for angular layers and show that, in a one-layer planted model, a sufficiently small \emph{G}-gradient can guarantee exact recovery of a target binary solution. We further prove uniform approximation and convergence guarantees that provide a rigorous path from tractable continuous optimization to exact recovery in the original discrete model. Empirically, we instantiate these ideas in \textit{Angular Nets}, including low-precision variants of ResNet, Vision Transformers, and BERT-style encoders, and obtain strong performance across vision and language tasks at substantially reduced precision. Overall, our results suggest that preserving angular structure provides a principled foundation for low-bit deep learning. Our full implementation is available at: \url{https://github.com/GGM2026/GGM}.

While recent vision-language models (VLMs) have achieved significant improvements on static visual-to-code tasks such as generating code for webpages, charts, or SVGs, it is still unclear VLMs are able to recover temporal dynamics when motion is present. To this end, we introduce the animation-to-code task, as well as a benchmark, \texttt{Animation2Code}, for evaluating temporal visual reasoning via reconstructing executable web animation code from videos. \texttt{Animation2Code} consists of 1,069 web animation videos with diverse visuals and motion patterns, augmented with corresponding HTML/CSS/JavaScript implementations. We further pair the dataset with several novel human-aligned metrics, which allow us to disentangle visual similarity from temporal alignment when comparing rendered animations to ground-truth samples. We benchmark current SOTA model performance on these new samples, and show that current models struggle with temporal consistency, even when achieving high appearance similarity, even in fine-tuning and iterative refinement scenarios. Our benchmark dashboard is available at: \url{https://anim2code-dashboard.vercel.app}.

Currently, evaluation of hallucination detection methods for vision-language models (VLMs) only reports aggregated statistics, overlooking potential confounding factors and providing limited insight into the strengths and weaknesses of each method. We conduct an in-depth analysis of what drives detection performance, focusing on methods that use either internal model signals (white box) or output logits (black box). First, our analyses show that a naive baseline of a word's token position in the caption is a competitive predictor of hallucinations and that many white box methods, despite performing the best, rely on this heuristic. Second, we find that some white box methods are additionally specialized to VLM architecture, such that when used with certain VLMs, they can even detect hallucinations in captions generated by other VLMs. Lastly, we find that the discriminative advantage of white box methods over black box primarily arises from detecting language-based hallucinations, as opposed to vision-based. Taken together, these analyses reveal insights into hallucination detection methods that are not captured by current evaluation protocols. Drawing from our findings, we encourage future work to develop more comprehensive evaluations that can better reflect hallucination detection behavior.

We present AnyEdit, a unified framework for speech and singing voice editing in real-world acoustic environments. Recent unified generative models already achieve strong speech and singing generation in clean conditions, but real-world vocal editing remains challenging because the model must revise content while simultaneously preserving speaker identity, prosody, and the surrounding acoustic scene. A straightforward solution is to directly model noisy edited audio end to end, yet this entangles semantic editing with environmental acoustics and can weaken the core vocal generation capability. We therefore adopt a two-stage design that preserves clean-domain editing ability while deferring acoustic rendering to a separate stage. First, an instruction-guided autoregressive editor predicts clean prosody and content-style tokens. To make this stage robust to degraded inputs, we introduce a teacher-student noise-robust prosody tokenizer that distills clean prosodic tokens from noisy recordings. Second, a flow-matching acoustic model performs in-context acoustic rendering conditioned on the source acoustic context, enabling the edited segment to inherit surrounding noise, reverberation, and background sound while maintaining speaker timbre consistently. To further stabilize the unedited regions, we introduce a unmodified-area-aware supervision mechanism. After pre-training, we also apply direct preference optimization (DPO) using real singing editing pairs mined via dynamic time warping to achieve better singing editing quality. To enable realistic evaluation, we introduce AnyEditBench, the first benchmark covering both speech and singing voice editing under diverse acoustic conditions. Experiments on Chinese and English speech and singing datasets showed that AnyEdit consistently outperforms strong baselines in content accuracy, subjective quality, and background consistency under noisy and reverberant conditions. Demo audios can be found at https://any-edit.github.io.

Conditional human motion generation remains a fundamental challenge in computer vision and robotics. Despite significant progress, current methods are often constrained by fixed modality configurations and task-specific architectures, leaving cross-modal interactions and the scaling laws of multimodal-conditioned synthesis largely underexplored. A key bottleneck is the scarcity of large-scale modality-aligned motion data, limiting generalization across diverse control signals. In this work, we introduce \textbf{OmniHuMo}, a large-scale, high-quality dataset comprising over 5,000 hours of motion and 3.2 million sequences with precisely aligned multimodal annotations (e.g., text, speech, music, and trajectory). Leveraging OmniHuMo, we propose \textbf{AnyMo}, a unified multimodal framework combining a Residual FSQ-based motion tokenizer with a scalable masked modeling transformer, enabling high-quality motion synthesis under arbitrary modality combinations. Extensive experiments show that AnyMo achieves high-fidelity synthesis while offering flexible control over both spatial and stylistic attributes.

Test-time scaling systems often choose how much to sample, which prompt to use, and which verifier or aggregation rule to trust after repeatedly inspecting a calibration set. Standard validation error bars are not valid after this adaptive selection. We model a test-time scaling recipe as a randomized predictor with a random cost and prove a time-uniform PAC-Bayes certificate that holds simultaneously over all calibration times, all posterior mixtures over recipes, and all data-dependent stopping rules. Thus the same event covers the recipe search that precedes deployment, not merely the fixed recipe finally reported. For a finite grid of budgets and verifiers, the penalty is the familiar $\log M/n$, while non-uniform priors and localized posterior mixtures give sharper certificates for structured searches. We specialize the result to self-consistency and show a complementary margin law: plurality voting improves exponentially in the sample budget only on examples where the correct answer is the generator's unique mode; otherwise more compute can provably amplify errors. The same margin analysis yields an oracle water-filling law for compute allocation. A synthetic audit quantifies the optimism caused by adaptive budget selection, and a 192-example GSM8K audit using gpt-4.1-mini generations illustrates risk-cost reporting for a real inference pipeline. The results give a lightweight, distribution-free way to report risk and compute claims for adaptive inference systems.

Recent progress in video generation has been remarkable, yet state-of-the-art models remain largely confined to 720p, well short of the 4K resolution of modern displays. Scaling video generation to higher resolutions is constrained by the scarcity of high-resolution video data and the prohibitive cost of large-scale training. Beyond data and compute, higher resolutions place substantially greater demands on spatial modeling, particularly for fine-grained details, complex compositions, and local structures. Such high-resolution spatial supervision, however, is abundantly available in high-resolution images, which are easier to collect and train on at scale. This naturally raises a question: can image-only supervision unlock native 4K video generation? To answer this, we propose ApertureAttn, an inverted attention pyramid that concentrates adaptation on high-resolution fine-grained spatial modeling using only single-frame high-resolution images, without any video training data. We further introduce window-scaled extrapolation and proximity-guided temporal attention to enhance spatiotemporal consistency. Extensive experiments show that our method enables native 4K video generation with strong visual fidelity and temporal coherence, while requiring only image supervision and minimal training cost. Despite this lightweight training setup, it remains competitive with state-of-the-art high-resolution video generation methods trained directly on high-resolution video data at substantially higher training cost.

Sequence models in machine learning compress long input histories into finite-dimensional states, from recurrent networks and reservoir computing to recent state space models for long-context data. This raises a basic expressivity question: how simple can the recurrent memory be before it can no longer approximate stable causal sequence-to-sequence maps? We answer this question for fading-memory operators, where the distant past has uniformly diminishing influence. We show that every continuous causal time-equivariant fading-memory operator on a compact input set can be uniformly approximated by a state space model whose recurrence is a fixed diagonal linear contraction. The dynamics are time-invariant, input-independent, and contain no nonlinear hidden-state update or delay-line structure. The construction uses a bank of short exponential traces; although such traces do not exactly store a recent input window, a singular Vandermonde extraction approximately reconstructs the window while the remaining tail becomes negligible as delta decreases. We also prove that exact delay recovery is impossible for any fixed finite trace bank, showing that the accuracy-dependent extraction is essential. Technically, the result separates recurrent expressivity from exact memory storage, while exposing a conditioning cost in the readout. More broadly, it clarifies why highly stable, structurally simple SSM cores can still be universal approximators for stable sequence behavior.

Unlearnable Examples (UEs) aim to protect data from unauthorized model training by injecting imperceptible perturbations that degrade model generalization. Existing methods typically rely on well-trained surrogate models to generate such perturbations, implicitly assuming that stronger surrogates yield better degradation. In this work, we challenge this widely adopted assumption and show that it is fundamentally suboptimal. We provide theoretical insights by establishing an upper bound on the target model's test loss, showing that perturbation effectiveness is closely related to the surrogate’s sensitivity to input perturbations. Specifically, a less robust surrogate model leads to stronger poisoning performance. Motivated by this analysis, we propose a simple yet effective stability metric to select optimal surrogate models for single-level methods. We further extend this principle to bi-level optimization frameworks. Beyond revealing the bi-level method's implicit reliance on non-robust surrogates, we substantially boost its poisoning performance by the stability-based surrogate selection as well. Experiments on five datasets and four representative UE methods show that our selected surrogates reduce the target model's test accuracy by an average of 18\% compared with well-trained surrogates. Notably, this is the first work to reduce CIFAR-10 test accuracy to 10\% under a perturbation budget of 1/255, whereas prior UE methods typically rely on 8/255.

Hierarchical reasoning model (HRM) achieves extraordinary performance on various reasoning tasks, significantly outperforming large language model-based reasoners. To understand the strengths and potential failure modes of HRM, we conduct a mechanistic study on its reasoning patterns and find three surprising facts: (a) Failure of extremely simple puzzles, e.g., HRM can fail on a puzzle with only one unknown cell. We attribute this failure to the violation of the fixed point property, a fundamental assumption of HRM. (b) "Grokking" dynamics in reasoning steps, i.e., the answer is not improved uniformly, but instead there is a critical reasoning step that suddenly makes the answer correct; (c) Existence of multiple fixed points. HRM "guesses" the first fixed point, which could be incorrect, and gets trapped there for a while or forever. All facts imply that HRM appears to be "guessing" instead of "reasoning". Leveraging this "guessing" picture, we propose three strategies to scale HRM's guesses: data augmentation (scaling the quality of guesses), input perturbation (scaling the number of guesses by leveraging inference randomness), and model bootstrapping (scaling the number of guesses by leveraging training randomness). On the practical side, by combining all methods, we develop Augmented HRM, boosting accuracy on Sudoku-Extreme from 55.0% to 96.9%. On the scientific side, our analysis provides new insights into how reasoning models "reason".

Vision Transformers (ViTs) face severe computational bottlenecks due to the quadratic complexity of self-attention at high resolutions. Existing token reduction methods rely on local metrics—such as single-layer attention scores—that are inherently vulnerable to the \emph{attention sink} phenomenon, where uninformative tokens are paradoxically preserved over salient foreground objects. We propose \textbf{ASAP} (Attention Sink Anchored Pruning), a training-free framework that recasts this sink as a feature. Modeling ViT information flow as a Lazy Random Walk, ASAP identifies the sink as a dominant accumulator of probability mass. By computing the \emph{diffusion distance} to the sink within the cumulative transition matrix, ASAP partitions tokens via \emph{Radial Diffusion Clustering} and compresses background redundancy through \emph{Transition Weight Pooling} in a single shot. Extensive experiments across image, video, and vision-language tasks demonstrate ASAP outperforms state-of-the-art methods, accelerating throughput by up to 48\% while maintaining—or even exceeding—baseline accuracy.

Virtual sensors replace expensive physical sensors in critical applications through machine learning by predicting target signals from available measurements. Existing virtual sensor approaches require application-specific models with hand-selected inputs for each sensor, cannot leverage task synergies, and lack consistent benchmarks. While emerging time series foundation models offer general-purpose, pretrained solutions in other domains, they are computationally expensive and limited to predicting their input signals, making them incompatible with virtual sensors. We introduce the first multi-task model for virtual sensors addressing both limitations. Our unified model can simultaneously predict diverse virtual sensors exploiting synergies while maintaining computational efficiency. It learns relevant input signals for each virtual sensor, eliminating expert knowledge requirements while adding explainability. In our large-scale evaluation on three standard benchmarks and an application-specific dataset with over 18 billion samples, our architecture reduces computation time by up to 415x and memory requirements by 951x, while maintaining or even improving predictive quality compared to unified baselines. Compared to existing isolated models for a single virtual sensor, our unified approach generates superior predictions at similar inference speed while scaling gracefully to hundreds of virtual sensors with nearly constant parameter count, enabling practical deployment in large-scale sensor networks.

Modern large language models are able to compose skills in order to perform complex tasks, many of which might not have been seen during training. The details of how exactly this composition occurs remain elusive. In this paper, we study a mechanism for compositional generalization in transformers by considering a simple controlled setting involving variable assignment and modular addition. By partitioning our training data into disjoint sets, we observe that small transformers are able to generalize to previously unseen combinations of variables and numbers. Our mechanistic analysis shows that the same ``modular addition'' MLP module is used whether the inputs are given directly or indirectly through a separate variable assignment mechanism. We also analyze the training dynamics from an empirical lens, which reveals three phases of learning: first, modular addition is learned, then the structure required for variable assignment, and finally a refinement phase where the model generalizes to sequences not seen in training. Finally, we provide a theoretical framework to explain how compositionality emerges from training dynamics. These results suggest that compositional generalization can be a natural consequence of the compositionality of internal mechanisms in transformers.

We study a stochastic bandit problem in which learning is split between two agents. In each round, a human observes rewards but can choose only between a pair of arms selected by an assistant, while the assistant observes the human's behavior but not realized rewards. We call this model the **assistive dueling bandit**. The human is modeled as a learning agent whose regret on any subset of arms is bounded by a function $g(T)$ unknown to the assistant. This is a natural model for recommender systems or AI assistants that must present a slate of options to a user who may be still learning about their own preferences. We provide a general reduction from assistive dueling bandits to the problem of **max-finding with imprecise feedback.** With it, we design assistant algorithms that cooperate with any $g(T)$-regret human to achieve a joint regret of $\tilde{\mathcal{O}}(K \cdot g(T/K))$, even without knowledge of $g(T)$. Crucially, when $g(T) \in \mathcal{O}(\sqrt{T})$, our method recovers the nearly optimal minimax rate of $\tilde{\mathcal{O}}(\sqrt{KT})$, implying that splitting up the responsibilities of learning in this way can be done without suffering any excess regret asymptotically. We complement our theoretical analysis with experimental evidence that this algorithm outperforms an assistant implemented via standard dueling bandit algorithms.

Flow-based generation in high-dimensional spaces is difficult because velocity prediction requires modeling high-dimensional noise, even when data has strong low-rank structure. We present Asymmetric Flow Modeling (AsymFlow), a rank-asymmetric velocity parameterization that restricts noise prediction to a low-rank subspace while keeping data prediction full-dimensional. From this asymmetric prediction, AsymFlow analytically recovers the full-dimensional velocity without changing the network architecture or training/sampling procedures. On ImageNet 256$\times$256, AsymFlow achieves a leading 1.57 FID, outperforming prior DiT/JiT-like pixel diffusion models by a large margin. AsymFlow also provides the first-ever route for finetuning pretrained latent flow models into pixel-space models: a patch-level linear lift initializes a low-rank pixel model whose denoising trajectory preserves the latent model's high-level semantics and structure, so finetuning mainly improves low-level residuals rather than relearning pixel generation. We show that the pixel AsymFlow model finetuned from FLUX.2 klein 9B establishes a new state of the art for pixel-space text-to-image generation, beating its latent base on HPSv3, DPG-Bench, and GenEval while qualitatively showing substantially improved visual realism. Code and models will be released publicly.

Audio--visual joint representation learning under Cross-Modal Generalization (CMG) aims to transfer knowledge from a labeled source modality to an unlabeled target modality through a unified discrete representation space. Existing symmetric frameworks often suffer from information allocation ambiguity, where the absence of structural inductive bias leads to semantic--specific leakage across modalities. We propose Asymmetric Hierarchical Anchoring (AHA), which enforces directional information allocation by designating a structured semantic anchor within a shared hierarchy. In our instantiation, we exploit the hierarchical discrete representations induced by audio Residual Vector Quantization (RVQ) to guide video feature distillation into a shared semantic space. To ensure representational purity, we replace fragile mutual information estimators with a GRL-based adversarial decoupler that explicitly suppresses semantic leakage in modality-specific branches, and introduce Local Sliding Alignment (LSA) to encourage fine-grained temporal alignment across modalities. Extensive experiments on AVE and AVVP benchmarks demonstrate that AHA consistently outperforms symmetric baselines in cross-modal transfer. Additional analyses on talking-face disentanglement experiment further validate that the learned representations exhibit improved semantic consistency and disentanglement, indicating the broader applicability of the proposed framework.

Autoregressive language models (ARMs) suffer from the reversal curse: after learning "$A$ is $B$," they often fail on the reverse query "$B$ is $A$." Masked diffusion language models (MDMs) exhibit this failure in a much weaker form, but the underlying reason has remained unclear. A common explanation attributes this mitigation to their any-order masked training objective. However, observing "$[\textnormal{\textbf{M}}]$ is $B$" during training teaches recovery of $A$ from $B$ in one positional configuration, and does not by itself explain why the learned evidence should transfer to the reverse prompt "$B$ is $[\textnormal{\textbf{M}}]$." We provide a theoretical analysis showing that this transfer arises from a parameter-level coupling between forward and reverse positional conditionals: shared Transformer parameters store token-pair evidence, while relative positional encodings route attention through queries and keys without changing the value-side evidence being retrieved. In a one-layer MDM, we prove that forward masked training strengthens evidence that is reusable in reverse queries, induces correlated forward--reverse attention routes, and yields a positively aligned shared-storage gradient component that decreases the reverse loss to first order. Controlled one-layer experiments and large-scale LLaDA/Dream experiments verify these signatures and show that they translate into improved reverse prediction.

Single-cell multi-omics technologies provide a powerful basis for characterizing cell-state transitions and dynamic regulatory processes. However, existing methods for multi-omics dynamic modeling still face two major limitations. From the data perspective, many methods require high-quality paired multi-omics measurements; from the modeling perspective, dynamic inference often relies on predefined regulatory structures or kinetic assumptions. These limitations restrict their applicability to partially paired, unpaired, and broader cross-omics settings. To address these challenges, we propose ATLAS, a unified framework for single-cell multi-omics alignment and dynamic modeling that jointly learns cross-omics consistency and temporal dynamics from partially paired or unpaired data. ATLAS adaptively models temporal lag effects between omics layers to characterize asynchronous cross-omics regulation, and further introduces a reliability-guided temporal distillation strategy to improve model-based temporal ordering. We systematically evaluate ATLAS on five datasets across four tasks, showing strong overall performance in multi-omics alignment, cross-omics prediction, trajectory inference, and future-state prediction. Our code is available at https://github.com/anomity/ATLAS.

Link prediction (LP) is a widely applied task in graph learning. Conventional LP approaches typically follow a dataset-specific paradigm, requiring independent training for each graph and incurring high computational and maintenance costs in large-scale applications. Motivated by these limitations, recent work explores Universal Link Prediction (ULP), aiming to enable training-free inference on arbitrary unseen graphs. However, existing methods primarily rely on subgraph sampling strategies to construct universal representations, which are unable to capture higher-order information due to the exponential growth of higher-order neighborhoods. As a result, they often under-perform on graphs where long-range signals are essential, such as biological networks. Moreover, their prediction models are typically domain-agnostic and lack the ability to adapt to different graph domains, leading to limited generalization. To address these challenges, we propose an Atlas-based Universal Link Prediction framework (AtlasULP), which introduces the relation atlas to capture structural associations between node pairs from both global and local perspectives, enabling a comprehensive characterization of connectivity signals across diverse graph domains. Building upon this representation, we develop a domain-guided prompting ULP model, which generates domain-aware prompt tokens from contextual structures and performs prompt-guided in-context prediction for adaptive link inference. Extensive experiments demonstrate that AtlasULP consistently outperforms state-of-the-art methods across diverse graph domains.

An attacker that strategically chooses when to attack is much harder to catch than one that attacks indiscriminately. AI control is a safety framework for deploying capable but untrusted AI agents under oversight from a weaker trusted monitor and a limited human audit budget. Control evaluations stress-test these protocols by pitting a red-team attack policy against the blue-team monitor, but current evaluations typically assume attackers that do not strategically select when to attack. However, a capable attacker would not attack indiscriminately. We study this capability, attack selection, in agentic settings by decomposing attack decisions into a start policy, which decides when an attacker should attack, and a stop policy, which decides when an attacker should abort an ongoing attack. Across two agentic settings, BashArena and LinuxArena, both policies substantially lower measured empirical safety without changing the underlying attack capability. At a 1% audit budget, our start policy reduces safety by 20pp on both BashArena and LinuxArena, and our stop policy reduces safety by 20pp on BashArena and 28pp on LinuxArena. These reductions in safety should be interpreted as upper bounds on the effect of attack selection. Existing control evaluations may therefore yield overly optimistic safety estimates against selective attackers. We recommend that future evaluations, system cards, and safety cases should elicit attack selection to produce more realistic safety estimates.

Speculative decoding accelerates LLM inference by drafting future tokens with a small model, but drafter models degrade sharply under template perturbation and long-context inputs. We identify a previously-unreported phenomenon we call \textbf{attention drift}: as the drafter generates successive tokens within a speculation chain, attention progressively moves from the prompt onto its own recently-generated tokens. We observe this across both \emph{EAGLE3} drafters and \emph{MTP heads}, suggesting drift is a property of drafter designs. We trace this to the un-normalized residual path between chain steps: the drafter's hidden state magnitude grows monotonically with chain depth, which exhibits dynamics consistent with additional pre-norm transformer layers stacked on the target rather than as a standalone autoregressive predictor. In order to limit the growth, we propose two architectural changes: Post-norm on the drafter hidden states and per-hidden-state RMSNorm after capturing target hidden states. Our interventions improve acceptance length over the current leading model, pre-norm EAGLE3, by up to $2\times$ under template perturbation, $1.18\times$ on long-context tasks, and $1.10\times$ on seven standard benchmarks spanning multi-turn chat, math, and coding. Our changes also allow shorter train-time-test depths to generalize over longer drafting sequences.

Protecting the intellectual property of open-weight large language models (LLMs) requires verifying whether a suspect model is derived from a victim model despite common laundering operations such as fine-tuning (including PPO/DPO), pruning/compression, and model merging. We propose AttnDiff, a data-efficient white-box framework that extracts fingerprints from models via intrinsic information-routing behavior. AttnDiff probes minimally edited prompt pairs that induce controlled semantic conflicts, captures differential attention patterns, summarizes them with compact spectral descriptors, and compares models using CKA. Across Llama-2/3 and Qwen2.5 (3B--14B) and additional open-source families, it yields high similarity for related derivatives while separating unrelated model families (e.g., >0.98 vs. <0.22 with M=60 probes). With 5--60 multi-domain probes, it supports practical provenance verification and accountability; our open-source implementation is available at https://anonymous.4open.science/r/AttnDiff-37A1/.

Speech-to-speech (S2S) language models are increasingly deployed as voice agents that schedule appointments, take grocery orders, and plan events. Reliable deployment in such settings requires accurate listening, grounded tool use, state tracking, knowledge-base access, and recovery from user corrections across many turns. Existing voice agent benchmarks often score final database state or per-call tool accuracy, which can obscure the first failure point when an early mistake cascades through the rest of a conversation. We introduce AudioAgentBench, a fixed-trace benchmark suite with six multi-turn voice-agent tasks, 221 turns of pre-recorded audio, golden tool calls, tool schemas, knowledge bases, and matched TTS and human-recorded variants. The fixed audio traces make turn-level expectations reproducible and support a diagnostic scoring protocol that attributes failures across tool use, state tracking, ambiguity handling, knowledge grounding, and instruction following. An LLM judge applies the protocol to each turn by comparing model outputs against golden tool calls, task state, and task-specific knowledge bases, while using cross-turn realignment, conditional penalty absorption, and category-aware dimension gating to avoid double-counting cascading errors. We evaluate seven S2S models across 420 continuous session runs. The top model averages an 80.3\% pass rate, but the top four models have overlapping 95\% confidence intervals, and no model leads on more than two of six tasks. On a 408-run turn-level diagnostic subset with 14,907 judged turns, tool-use accuracy on turns where a tool call was expected peaks at only 61.4\% and drops to 32.3\% on a confusable-name scheduling task. Errors cascade by up to 3.15× in stateful tasks, and tool-use accuracy on a 30-turn grocery task falls by 40.5 percentage points from the first half to the last half. Scoring only turns where a tool call was expected reduces apparent tool-use accuracy by 22–44 percentage points across models, which reveals a larger gap between strong and weak systems than aggregate task success suggests. AudioAgentBench surfaces action, state, and recovery failures that final success metrics obscure, providing a finer-grained diagnostic benchmark for production-oriented voice agents.

In-context learning (ICL) allows large language models (LLMs) to adapt to new tasks from a few examples without parameter updates. Previous work and our empirical studies suggest two modes in ICL: Task Recognition, which applies pre-trained knowledge to familiar tasks, and Task Learning, which generalizes to novel mappings during inference. However, a unified theoretical account of these behaviors remains lacking. We propose a new Bayesian prediction framework that models pre-training data as a Pitman–Yor mixture over latent tasks, which captures the growing and heavy-tailed structure of natural language. By formulating an information-theoretic optimization problem, we derive the structure of the optimal ICL predictor under explicit capacity constraints. Our analysis reveals a phase transition in optimal capacity allocation: a water-filling strategy that prioritizes high-frequency tasks while assigning zero task-specific information to rare ones. This gives rise to two regimes: for frequent tasks, the model performs Task Recognition, yielding exponentially fast error decay with prompt length. For rare or unseen tasks, the model performs Task Learning by executing an implicit inference algorithm, resulting in a slower power-law convergence rate. Our results provide a principled explanation for the dual nature of ICL and establish a direct connection between pre-training data distributions, model capacity, and in-context generalization behavior.

Professional image retouching is a sequential and reasoning-driven process. Photographers first analyze scene semantics, lighting structure, and depth cues to design a retouching plan, and then progressively apply global tone grading and mask-guided local corrections based on the evolving visual outcome. Recent MLLM-empowered retouching agents either predict global parameters or respond to user-provided retouching instructions. Although showing impressive results, they lack the capacity for autonomous sequential reasoning and fall short of fine-grained local control. We propose AURA, an AUtonomous Retouching Agent that emulates the workflow of expert photographers. Given an input photograph, AURA autonomously assesses the scene and proceeds progressively. At each step, it reasons about what to retouch, executes the adjustment, and feeds the result back as visual context for next-step reasoning, forming a fine-grained visual chain-of-thought that evolves with the image. To support AURA, we contribute the first expert-annotated long-horizon retouching trajectory dataset, where expert photographers edit the input image from scratch with deliberate artistic intent. Based on this dataset, we train AURA in two stages: supervised fine-tuning to learn the retouching trajectory, followed by GRPO-retouching, an agentic reinforcement learning stage that optimizes for perceptual quality and tool accuracy. Our experiments demonstrate that AURA consistently outperforms existing methods in both quantitative metrics and user studies, enabling one-click professional retouching and democratizing masterpiece previously accessible only to skilled experts.

Language agents increasingly operate over streams of related tasks, yet existing memory systems struggle to convert accumulated experience into reusable knowledge. Retrieval-augmented and structured memory methods record per-session observations effectively, but often couple acquisition and consolidation into a single online process, leaving the agent without a global view across sessions to discover recurring patterns, abstract shared procedures, or prune redundant entries. Inspired by complementary learning systems theory, we propose Auto-Dreamer, a learned offline consolidator for language-agent memory. Auto-Dreamer decouples fast per-session memory acquisition from slow cross-session consolidation. Given a region of a typed memory bank, the consolidator performs bounded tool-use to search memory, inspect candidate entries, and trace them back to raw source trajectories, synthesizing provenance-grounded replacement memories that supersede the original region. We train Auto-Dreamer via GRPO, using end-to-end agent performance as the reward signal to learn how to consolidate memories acquired through fast online experience. Trained on ScienceWorld trajectories alone, Auto-Dreamer transfers without retraining to held-out ALFWorld and WebArena, improves task success over fixed, RL-trained, and prompted memory baselines in both continual-memory deployment and fixed-bank consolidation, and does so with an active memory bank an order of magnitude smaller than competitive baselines.

Causal inference is central to scientific discovery, yet choosing appropriate methods remains challenging due to the complexity of statistical methodology and real-world data. Inspired by the success of artificial intelligence in accelerating scientific software, we introduce an evolutionary framework that uses large language models to discover and iteratively refine causal methods. Across benchmarks, our estimators consistently outperform established baselines: our best estimator lay on the Pareto frontier of 58 human submissions for a recent community competition. We also extend the algorithm to achieve competitive results in settings with estimated rewards. Analysis of the evolutionary trajectories shows that agents progressively discover sophisticated strategies tailored to unrevealed data-generating mechanisms. Our findings suggest that language-model-guided evolution could be used in scientific settings with partially observed rewards such as causal inference.

Emotions are expressed on a continuum and are naturally described in the valence-arousal (VA) space. Existing VA-based image editors are typically training-based and domain-specific, limiting them to VA regions seen during training and making them brittle across datasets and domains. Meanwhile, despite producing high-quality edits, state-of-the-art foundation image editors are not continuous affective controllers: non-linear emotion representations in text encoders makes prompt interpolation poorly calibrated, so accurate and continuous VA control remains unreliable. To fill this gap, we introduce Barycentric Guidance, a training-free framework that turns pretrained foundation image editors into continuous affective controllers over VA, requiring no finetuning, extra data, or model changes. Our method maps any target VA point to simplex-based guidance weights, enabling VA controllability in three aspects: (i) accurate targeting of VA states, (ii) broad coverage across the VA plane, and (iii) continuous control. Across three face datasets, we improve VA target accuracy by 25–29\%, expand VA-plane coverage by 28–50\%, increase expression diversity by 29–49\%, over the strongest baseline. Unlike prior domain-specific methods, our training-free approach exploits the foundation model’s inherent versatility, enabling generalization beyond human faces to animals, artworks, and complex scenes within a single framework. Code will be made publicly available.

Block-diffusion drafters have recently shown strong potential for speculative decoding by predicting multiple future-token distributions in a single forward pass. To exploit these parallel predictions, tree-based verification can check multiple plausible continuations instead of committing to a single greedy draft path, often increasing the number of accepted tokens per decoding cycle. However, existing tree-based approaches typically rely on fixed tree topologies or static verification budgets, even though the best budget depends on the local draft distribution, context length, target model, and hardware runtime. We propose a system-aware adaptive tree framework for block-diffusion speculative decoding. At each decoding cycle, our method builds a tree toward high-probability draft paths and selects the verification budget that maximizes an online speedup estimate. This estimate combines a drafter-based acceptance surrogate with an analytical verifier-latency model calibrated from observed runtimes. The framework requires no additional training of the drafter or target model and preserves the target model's decoding rule. Across target models, benchmarks, decoding temperatures, and GPU platforms, our method achieves up to a $6.93\times$ wall-clock speedup over standard autoregressive decoding. This corresponds to approximately 40% higher speedup than Dflash, the strongest existing block-diffusion drafter, and is obtained without per-setting budget tuning.

Feed-forward models for 3D reconstruction have achieved strong performance using deep cross-view attention to exchange information across images. However, these approaches often depend on heavy decoder stacks and lack a structured mechanism for geometry refinement, resulting in poor multi-view consistency. We address this by drawing inspiration from classical bundle adjustment (BA), which can be viewed as an iterative information propagation process between poses and local geometry. Inspired by BA, we propose BA-T, an iterative Transformer that implements BA-style structured updates as a repeatable layer in implicit token space. Instead of relying on deep attention stacks, BA-T refines predictions based on latent residual by a single lightweight layer. Experiments demonstrate that BA-T progressively improves pose and reconstruction accuracy across iterations, achieves stronger cross-view consistency than conventional decoders, and matches or surpasses substantially larger models while using only 16\% of their decoder parameters. BA-T provides a compact, efficient, and structural alternative to depth-heavy attention, enabling accurate 3D reconstruction within a lightweight architecture. The code will be made publicly.

Existing semi-supervised adversarial training (SSAT) methods typically employ a teacher-student framework where a teacher model provides supervisory signals for unlabeled data. However, they rely on static supervision---either hard pseudo-labels or soft labels with fixed temperature---which maintains constant learning difficulty throughout training. This rigidity fails to accommodate the student's evolving capability, especially when the teacher's performance plateaus. To address this, we reframe SSAT as knowledge distillation from a Bayes teacher and formalize a stage-dependent bias-variance tradeoff. This analysis reveals the existence of an optimal temperature: low temperatures reduce bias, while higher temperatures control the variance term in the robust generalization bound. Guided by this insight, we propose BayesAT, a Bayes-guided progressive distillation framework that jointly employs a low-to-high temperature warming schedule and confidence-adaptive sample reweighting. BayesAT allows the student to first learn from sharp, high-confidence supervision to establish reliable decision boundaries, then progressively from smoother distributions that encourage exploration of inter-class relationships. Extensive experiments on CIFAR-10, CIFAR-100, and ImageNet-200 demonstrate that BayesAT consistently improves robust accuracy while maintaining natural accuracy over existing SSAT methods, highlighting the importance of dynamic, theory-driven supervision for effective knowledge transfer in semi-supervised adversarial training.

Causal experimental design often assumes direct control over treatment assignment, but many experiments can only assign encouragements that affect treatment receipt indirectly. We formulate this as adaptive instrumental variable (IV) design under noncompliance: the experimenter chooses unit--instrument queries, observes stochastic treatment realizations and outcomes, and seeks the treatment-level conditional average treatment effect (CATE). This creates an acquisition mismatch: uncertainty in prospective feedback does not necessarily correspond to uncertainty in the target causal estimand, so outcome-predictive or marginal variance-based criteria may select queries weakly informative for CATE. We introduce instrumental variable information gain (IVIG), a Bayesian acquisition principle that scores queries by the expected information their joint treatment-realization and outcome feedback provides about CATE values over a target population. Under the working Bayesian posterior, IVIG is Bayes-risk aligned under logarithmic loss; we also characterize the residual information discarded by outcome-only acquisition and connect greedy IVIG-style acquisition to target posterior-variance reduction under a fixed-covariance analysis. To approximate IVIG, we use an empirical-Bayes Gaussian Process IV posterior approximation with first-stage treatment-realization modeling and treatment-realization-specific fantasy updates. Experiments on synthetic and two semi-synthetic IV benchmarks show improved CATE sample efficiency over adaptive-design baselines; a real-label diagnostic shows improved recovery of full-data CATE estimates from observed IV data.

Standard graph neural networks (GNNs) are built on message passing and neighborhood aggregation, following the Weisfeiler--Lehman refinement paradigm. Despite their wide adoption, these models tend to compress and mix neighbor information from the very first layer. This early mixing obscures the critical subgraphs that drive the prediction. Attention-based and subgraph-based methods partially address this by reweighting nodes or enriching local representations, but both families remain passive as they usually rely on soft weighting or predefined subgraph templates. We propose BeamWitness, a selective graph representation learning framework that treats subgraph identification as an active search problem. Starting from a root node, a policy trained with task rewards expands a connected subgraph one node at a time or terminates early, and beam search retains a small set of candidate subgraphs as witnesses for prediction. For node-level tasks, the target node serves as the root, while for graph-level tasks, BeamWitness launches searches from multiple root nodes and aggregates the selected witnesses into a graph-level representation. On controlled motif benchmarks with known ground-truth subgraphs, BeamWitness achieves near-perfect accuracy and selects witnesses that align with the true motifs. On real-world node and graph classification benchmarks, BeamWitness is competitive with standard GNNs and related methods while providing an explicit, interpretable subgraph-based account of each prediction.

We trace factual sycophancy across three post-training pipelines (OLMo 3 7B Think and Instruct, Llama 3.1 8B Instruct, Tulu 3) on AMPS and MedQuAD with a dual-track evaluation: a GPT-4o-judged generative track and a log-probability track reporting $\Delta$LogOdds separately under \emph{in-context} rebuttals and \emph{preemptive} wrong-answer assertions. Decomposing by challenge type $\times$ context reveals three patterns. (1) \textbf{Challenge type:} the sycophantic shift is alternative-conditioned, so simple pushback with no asserted answer produces no shift, while ethos/justification/citation challenges produce substantial positive shifts. (2) \textbf{Context:} every base model's sycophantic shift concentrates preemptively, with weak or defensive in-context response. On computational, IC diverges into four pipeline-specific endpoints. On medical, the same four pipelines converge with no defensive IC developing on any recipe. (3) \textbf{Behavioral vs log-probability:} post-training produces a \emph{surface-predictive dissociation}: matched-subset behavioral flip rate drops significantly on identical items while preemptive $\Delta$LogOdds on those same items grows.

Modern language models are commonly adapted after pretraining to follow instructions, align with user preferences, and improve deployment behavior. Such post-training often relies on preference data from users, annotators, or model interactions, which may contain sensitive prompts, private responses, proprietary tasks, or confidential judgments. Understanding the privacy implications of this data is therefore essential. Membership inference attacks (MIAs), which test whether a record was used for training, are the de facto standard for empirical privacy auditing in machine learning. However, preference-based post-training changes the unit of membership: a training record contains a prompt, a preferred response, and a dispreferred response, rather than a single input-output sequence. Audits that ignore this structure can under-report leakage. We therefore introduce a benchmark protocol that adapts strong reference-model MIAs to complete preference records. Our protocol asks whether membership is exposed by either response alone, by the model’s preference between them, or by the two responses jointly. We find that auditing the response pair can reveal membership evidence missed when the record is reduced to a single score. Across three preference datasets, two model families, and seven post-training objectives spanning three post-training families, measured leakage depends strongly on both the audited record statistic and the training objective, with imbalanced risk between preferred and dispreferred responses. We further evaluate the impact of parameter-efficient fine-tuning (PEFT) and differential privacy (DP), finding widely varying privacy-utility trade-offs. Overall, preference-based post-training leaks in ways that standard single-response audits can miss, motivating formal privacy methods that protect complete preference records while preserving post-training utility.

Reinforcement learning (RL) is now standard for post-training large language models. The same reward optimization that elicits useful capabilities, however, can also reinforce latent backdoors planted earlier in the pipeline: attack success rates under 0.2% after SFT rise to 40--98% on training-distribution inputs after RL, and generalize to 13--30\% on held-out evaluation, with no modification to the RL data, reward function, or training loop. We study this phenomenon in the context of agent models, where SFT poisoning teaches the victim model (Qwen3-8B and a small frontier model) to call an attacker-controlled oracle. Because the oracle can be set up to always return correct answers, calling it earns higher reward than the model's own attempts, and RL reinforces the behavior. We further show that the oracle's RL-time responses can instill persistent biases, such as brand preferences, that survive into deployment even when no tool calls happen. Such patterns appear difficult to detect with current tools: traditional guard models do not flag this mechanism, and LLM-based auditors remain unreliable, achieving only 4.5\% precision even after iterative prompt refinement. Our findings point to the value of post-RL safety evaluation, in particular scrutinizing tool-call patterns such as unnecessary external invocations.

Flow matching has recently emerged as a promising alternative to diffusion-based generative models, particularly for text-to-image generation. Although flow matching places no restriction on the source distribution, most existing systems still inherit a standard Gaussian from diffusion models, and the source is rarely treated as an optimization target at this scale. Recent works have begun to revisit this choice through condition-dependent or learned sources, yet evidence that such designs are effective in modern text-to-image systems—with high-dimensional latents and tightly integrated conditioning—remains limited. In this work, we study condition-dependent source distributions for flow matching along three axes: _why_ source learning helps, through the lens of the intrinsic variance term in the flow-matching objective; _how_ to make it work in modern text-to-image systems, where variance-only regularization and directional source—target alignment are critical for stable end-to-end training; and _when_ it is most beneficial, by connecting source design to recent representational advances in generative modeling and identifying target representation regimes in which learning the source yields the largest gains. Extensive experiments across multiple text-to-image benchmarks, backbones, and scales demonstrate that principled source design yields consistent and robust improvements—including up to $\mathbf{3.01\times}$ faster convergence in FID and $\mathbf{2.48\times}$ in CLIP score—and outperforms representative prior conditioned-source and condition-aware coupling methods.

Recent studies have shown that leveraging the reasoning and analytical capabilities of large language models (LLMs) for long video understanding has become a promising approach. However, these methods are constrained by a structural limitation: representing video as a flat frame sequence makes it difficult to model the temporal structure and logical dependencies of events, which in turn hampers long-range reasoning and risks overlooking important information. To address this limitation, we propose LVGraph, a framework that models video content using a coarse-to-fine hierarchical graph. LVGraph constructs a semantic hierarchy where high-level graphs characterize the global event structures within the video, while low-level graphs capture fine-grained, query-relevant information. Instead of linear scanning, reasoning is performed via a query-aware traversal of this graph, adaptively identifying and retrieving the most salient keyframes by navigating the nodes and edges most relevant to the query. Finally, we leverage a Vision Language Model (VLM) to augment the keyframe information, and subsequently answer the question by reasoning over the graph and the enriched keyframe captions. Extensive experiments on long video understanding benchmarks confirm the effectiveness of our method. LVGraph significantly outperforms existing LLM-based state-of-the-art methods on EgoSchema, NExT-QA, and long-duration segments (averaging 44 minutes) of the Video-MME benchmarks.

Early-exit networks promise inference savings by terminating computation at intermediate classifiers, but FLOPs-based gains rarely translate into on-device latency reduction under static-graph compilation. On NVIDIA Jetson Orin Nano, the per-sample exit policy of Shallow-Deep Networks (SDN) runs approximately 2× slower than vanilla ResNet-56 and misses the 60 fps deadline despite using only 40% of the FLOPs, because dynamic branching precludes layer fusion. We propose the Selective Lightweight IC Ensemble Network (SLIENet), a train-full, deploy-partial framework compiled as a single static FP16 engine. SLIENet trains the full backbone with light self-distillation, using the final classifier as a stop-gradient teacher to preserve error diversity across internal classifiers (ICs). At deployment, a sub-second calibration search selects an IC subset for softmax averaging, while overthinking late blocks are truncated at depth k. On CIFAR-100 across ResNet-56, VGG-16, and MobileNetV1, SLIENet outperforms SDN-based variants and remains competitive with ZTW. On Jetson Orin Nano, the recommended SLIE5 configuration improves accuracy by +2.46 pp while reducing p50 latency by 14% and energy by 15% over vanilla, with zero deadline misses across 10,000 test samples. Thus, multi-exit deployment should be a pre-compiled deterministic inference plan, not a sample-wise routing policy.

Current multivariate time series forecasting methods mainly rely on static linear decoders, but these often suffer from severe representational bottlenecks. In this paper, we propose a novel architecture called DecodeTS (\textbf{\underline{D}}ynamic \textbf{\underline{E}}xpert-\textbf{\underline{C}}oupled \textbf{\underline{O}}ptimal \textbf{\underline{DE}}coder for Time Series Forecasting), which replaces the conventional static prediction head with a heterogeneous expert library. DecoderTS adopts a divide-and-conquer strategy to disentangle complex temporal dynamics, such as long-term trends and abrupt changes. Crucially, we introduce an optimal transport (OT)-based dynamic routing mechanism that can adaptively assign customized combinations of experts to different variables. By imposing OT marginal constraints, DecoderTS is theoretically proven to achieve collaborative load balancing among experts and effectively eliminate representation collapse. Extensive experiments on more than \textbf{15} datasets and about \textbf{30} baselines demonstrate that DecoderTS achieves state-of-the-art forecasting accuracy, which delivers a \textbf{17.06\%} performance gain while attaining up to about \textbf{41×} inference time speedup and up to \textbf{3.61×} reduction in memory footprint.

Recent Autonomous Driving (AD) works such as GigaFlow and PufferDrive have unlocked Reinforcement Learning (RL) at scale as a training strategy for driving policies. Yet such policies remain disconnected from established benchmarks, leaving the performance of large-scale RL for driving on standardized evaluations unknown. We present BehaviorBench -- a comprehensive test suite that closes this gap along three axes: Evaluation, Complexity, and Behavior Diversity. In terms of Evaluation, we provide an interface connecting PufferDrive to nuPlan, which, for the first time, enables policies trained via RL at scale to be evaluated on an established planning benchmark for autonomous driving. Complementarily, we offer an evaluation framework that allows planners to be benchmarked directly inside the PufferDrive simulation, at a fraction of the time. Regarding Complexity, we observe that today's standardized benchmarks are so simple that near-perfect scores are achievable by straight lane following with collision checking. We extract a meaningful, interaction-rich split from the Waymo Open Motion Dataset (WOMD) on which strong performance is impossible without multi-agent reasoning. Lastly, we address Behavior Diversity. Existing benchmarks commonly evaluate planners against a single rule-based traffic model, the Intelligent Driver Model (IDM). We provide a diverse suite of interactive traffic agents to stress-test policies under heterogeneous behaviors, beyond just using IDM. Overall, our benchmarking analysis uncovers the following insight: despite learning interactive behaviors in an emergent manner, policies trained via pure self-play under standard reward functions overfit to their training opponents and fail to generalize to other traffic agent behaviors. Building on this observation, we propose a hybrid planner that combines a PPO policy with a rule-based planner, providing a baseline for our new benchmark. Code is available at https://anonymous.4open.science/r/behavior-bench-707C

Dense correspondence matching has historically been bounded by simplifying spatio-temporal priors, such as smooth motion and rigid geometry. While effective for classical tasks, these foundations collapse at the frontier of Vision-Language-guided Image Editing and Generation (VL-IEG), where transformations yield correspondences that are perceptually obvious yet physically discontinuous. To transcend spatio-temporal priors, we draw inspiration from human cognition: learning a generalized model of identity-preserving visual consistency rather than relying on physical constraints. To achieve this, we introduce FreeMatching, a unified framework that extracts identity-preserving knowledge from foundation models and diverse datasets. The model's capability is forged through a two-stage training paradigm: supervised pre-training on diverse annotated datasets, followed by weakly-supervised refinement on data without dense annotations. Experimentally, FreeMatching not only achieves performance competitive with SOTA methods on classical benchmarks but also establishes the first strong baseline for the challenging VL-IEG task. Furthermore, we demonstrate its utility as a quantitative metric for evaluating identity-preserving consistency in generative editing, aligning closely with human judgment.

Understanding how a prediction model will perform in a new environment before deployment is essential to preventing harm when algorithms inform decision-making. Two common sources of model performance degradation are (i) covariate shift, where the target covariate distribution differs from the source, and (ii) selective labels, where the observability of outcomes depends on historical decisions. We study pre-deployment model evaluation under the joint presence of covariate shift and labeling of outcomes selectively based on observed features. In particular, we present a double machine learning procedure for estimating the target risk of an arbitrary black-box prediction model under a general loss function. We show identification of this estimand under standard assumptions and derive a bias-corrected estimator based on the influence function of the target risk. Finally, we evaluate our estimator through experiments using the eICU electronic health records database, showing that it tracks the true target risk more accurately than methods that address either selective labels or covariate shift alone, as well as baselines that combine standard plug-in approaches.

Cell-level dense prediction is central to computational pathology, but remains challenging due to fine-grained histological structures, strong domain shifts, and costly dense annotations. Existing ViT-based pathology foundation models rely on patch tokenization, which can disrupt spatial continuity and weaken local morphological details needed for cell-level prediction. To address this, we propose Masked-Diffusion Convolutional Foundation Models, termed ConvNeXt Masked-Diffusion (CMD), a self-supervised convolutional generative pretraining framework for dense pathology representation learning. CMD uses a fully convolutional ConvNeXt-UNet backbone, performs masked-diffusion pretraining in pixel space, and incorporates frozen pathology foundation model features through adaptive normalization. Experimental results demonstrate that CMD consistently outperforms existing ViT-based pathology foundation models and even surpasses state-of-the-art end-to-end segmentation methods while fine-tuning only a small number of task-specific parameters across multiple pathology dense prediction tasks. The advantage is particularly pronounced under limited annotation settings, where CMD exhibits stronger robustness and generalization ability. Our findings suggest that purely convolutional architectures can also serve as competitive pathology foundation models for cell-level dense prediction, achieving leading performance within the current ViT-dominated paradigm and providing a scalable, high-performance solution that better preserves histological structural priors for fine-grained pathology understanding. Project page: https://anonymous.4open.science/r/ConvNeXt-Masked-Diffusion-5BC6.

EEG foundation models can learn generalizable representations from large-scale EEG corpora to enable single-backbone transfer across diverse clinical and brain-computer interface tasks. Existing models typically discretize the continuous multi-channel EEG waveform into patches or codebook tokens and train a transformer with masked self-supervision. Recognizing that this discretization fragments continuous brain rhythms and obscures fine-grained temporal dynamics, we present B[FM]$^2$ (Brain Foundation Model via Flow Matching), which drops discretization and pretrains directly on the raw signal using continuous-time flow matching without patches, tokenization, or masking. However, multi-channel EEG signals pose an architectural challenge for flow matching: time is densely sampled and highly autocorrelated (thousands of timepoints), while the electrode axis is short (tens of channels) at distinct scalp positions. To address this time-electrode asymmetry, we introduce SplitUNet, a velocity network that factorizes each block into separate 1D temporal and 1D electrode convolutions and downsamples only along time, preserving electrode topology throughout the hierarchy. B[FM]$^2$ sets a new state of the art on $7$ of $9$ standard downstream EEG classification tasks, using a pretraining budget of only $36{,}895$ segments ($\approx 307$\,h), a fraction ($\approx 3.3$\%) of that required by existing EEG foundation models. It also produces synthetic EEGs that two board-certified neurologists cannot distinguish from real EEGs (Cohen's $\kappa = -0.096$).

Diffusion probabilistic models have become a cornerstone of modern generative AI, yet the mechanisms underlying their generalization remain poorly understood. In fact, if these models were perfectly minimizing their training loss, they would just generate data belonging to their training set, i.e., memorize, as empirically found in the overparameterized regime. We revisit this view by showing that, in highly overparameterized diffusion models, generalization in natural data domains is progressively achieved during training before the onset of memorization. Our results, ranging from image to language diffusion models, systematically support the empirical law that memorization time is proportional to the dataset size, consistent with a kernel-regression bound on the time required to fit the empirical score at low noise. Generalization vs. memorization is then best understood as a competition between time scales. We show that this phenomenology is recovered in diffusion models learning a simple probabilistic context-free grammar with random rules, where generalization corresponds to the hierarchical acquisition of deeper grammar rules as training time grows, and the generalization cost of early stopping can be characterized. We summarize these results in a phase diagram. Overall, our results support that a principled early-stopping criterion – scaling with dataset size – can effectively optimize generalization while avoiding memorization, with direct implications for hyperparameter transfer and privacy-sensitive applications.

Post-training quantization (PTQ) to FP4 has emerged as a key technique for reducing large language model inference cost, especially on Blackwell GPUs which offer native FP4 tensor core support. Direct PTQ to FP4, however, still leaves a noticeable accuracy gap. Recent work on SVDQuant mitigates the errors introduced through FP4 quantization for diffusion models by adding high-precision low-rank corrections obtained via SVD. However, this high-precision low-rank correction requires many bits per rank, and moreover competes on tensor core utilization with the main FP4 GEMM. We introduce BiLoCo, a correction whose left and right factors are stored as binary ${\{\pm 1\}}$ sign vectors rather than as BF16 numbers. The signed format uses $16{\times}$ less storage per rank-one term, and since the correction now consists of additions/subtractions, this can be run on CUDA cores concurrently with the FP4 tensor-core GEMM. On 4--32B LLMs, BiLoCo matches or improves memory-matched SVDQuant for PTQ. We implement BiLoCo on B200 with a correctness-preserving CUDA schedule that runs end-to-end decode at $1.13$--$1.15\times$ FP4-only, with the residual gap coming from the final output combine rather than from one-bit arithmetic.

A language model agent acts on a context assembled by retrieval, summarisation, prompt templates, and memory. How many decision-relevant bits must this context carry before the agent can act with low regret? We give an answer that does not depend on the model, decoder, or scaffold. Modelling the agent as a Markov chain H→C→A and treating Bayes regret (the value gap between the chosen and the optimal action) as the distortion measure, we define a regret rate-distortion (RRD) function Rℓ(ε) whose converse is a one-line consequence of the data-processing inequality (DPI). A context that carries fewer than Rℓ(ε) bits cannot drive expected regret below ε, regardless of decoder, scaffold, scale, or chain-of-thought. For finite M-ary decisions under zero-one loss the bound reduces to the Fano floor in closed form, requiring ≈3.14 bits at (M=16, ε=0.1) and ≈6.73 bits at (M=256, ε=0.1). We instantiate the theory on 441 controlled single-step tool-selection cells, a balanced 16-way classification proxy spanning three open-source large language models (Qwen3-32B, Gemma3-27B, GLM-4-32B) and three public benchmarks (API-BANK, τ-bench, ToolBench). Because the gold tool index X=f(H) is a deterministic function of the history, the action-side rate IMM(X; A) lower-bounds the context capacity by data processing, so an action-side test of the converse is strictly tighter than a context-side one. Every cell respects the predicted floor (median margin +1.07 bits). Across 629 matched-rate pairs, 97.5% share overlapping regret confidence intervals (CIs), and a chain-of-thought supplement slides cells along the frontier rather than across it. The same data exposes a tokens-versus-bits gap reaching ∼1.7×10⁶:1, with the model extracting fewer than 0.05 bits about the gold action on a 32,768-token ToolBench prompt. In the single-step regime, the framework supplies a falsifiable, model-agnostic stopping criterion at IMM=Rℓ(ε) and reframes context engineering as an information-allocation problem.

We introduce BLARM, a feed-forward method for video-driven 3D mesh animation. Given a monocular video and a static object mesh, BLARM predicts a temporally coherent animated mesh whose motion follows the video. Rather than relying on explicit rigs or directly regressing high-dimensional vertex motion, we represent animation using a compact set of learned, time-varying rigid motion components and time-invariant vertex-to-component skinning weights. This yields a low-dimensional deformation space without requiring skeletons, cages, skinning weights, or rig annotations. Our architecture conditions geometry-derived deformation latents on video features through factorized spatial-temporal attention, then decodes rigid transformations blended by predicted skinning weights. Trained with trajectory reconstruction, entropy regularization, and motion-aware contrastive learning, BLARM produces accurate and temporally stable animations while recovering compact, interpretable motion structure from monocular video.

Tropical cyclone forecasting models are often trained and evaluated with reanalysis inputs. Reanalysis provides physically rich environmental context, but its most recent fields may be unavailable at forecast issuance. This can introduce a hindsight-style advantage and lead to optimistic estimates of real-time forecast skill. Satellite observations, in contrast, are available closer to real time but provide incomplete and noisy views of the storm system. We formulate this data-availability mismatch as forecasting through a reanalysis blind window. To study this setting, we introduce a real-time-faithful benchmark spanning 1980--2023 that combines lagged reanalysis with visible, infrared, water-vapor, and passive microwave satellite imagery. We then develop RECAST-TC, a multimodal reconstruction framework that fuses time-lagged environmental histories with real-time satellite observations to estimate a forecast-sufficient latent storm state from delayed, partial, and noisy inputs. We use idealized analysis-available forecasting and delayed-reanalysis-only forecasting as reference regimes. RECAST-TC consistently improves track and intensity forecasts over delayed-reanalysis baselines and moves noticeably closer to the idealized reference. These results highlight data availability and observation timing as central design variables for realistic evaluation and deployment-oriented AI forecasting of tropical cyclones.

Recently, reinforcement learning (RL) has been widely applied during post-training for diffusion large language models (dLLMs) to enhance reasoning with block-wise semi-autoregressive generation. Block size has therefore become a vital factor in dLLMs, since it determines the parallel decoding granularity and affects the rollout trajectories during RL optimisation, e.g., GRPO. Instead of investigating the effect of block size during inference on individual domains, this paper studies block size from a domain conflict perspective for dLLM RL post-training in multi-domain scenarios. The main contributions are: (1) a formulation of domain block size conflict in multi-domain RL for dLLMs, which will largely affect the post-training effectiveness for rollout-based RL methods; (2) a novel dataset, \textbf{Block-R1-41K} is constructed with a best-improved training block size for each sample, which also induces a Block Size Conflict Score to quantitatively measure the domain conflict; (3) a new benchmark, Block-R1, for flexible RL post-training for dLLMs in both single and cross domain; and (4) a simple yet powerful cross-domain post-training method with sample-level best-improved training block sizes. Extensive experiments on 13 distinct datasets, 7 latest RL algorithms, and various different dLLM backbones are covered in Block-R1. The benchmark is open-sourced at https://anonymous.4open.science/r/Block-R1-2026, with the dataset released at https://huggingface.co/datasets/dLLM-R1/Block-R1-41K.

Vector quantization is a fundamental primitive for scalable machine learning systems, enabling memory-efficient storage, fast retrieval, and compressed inference. Recent rotation-based quantizers such as EDEN, RabitQ, and TurboQuant have introduced strong guarantees and empirical performance, but the surrounding comparisons have been difficult to interpret because they rely on different distortion criteria, probability regimes, and implementation assumptions. As our first contribution, we provide a unified theoretical comparison of these methods and show that their relative advantages are criterion-dependent rather than absolute: TurboQuant is favorable for MSE distortion, EDEN is effective for expected inner-product distortion, and RabitQ provides strong high-probability control. This comparison clarifies the design principles behind each advantage and shows that no existing method uniformly dominates across the relevant measures. As our second contribution, we introduce Block-Sphere Quantization (BlockQuant), a new rotation-based block quantization algorithm designed around the spherical geometry of randomly rotated vectors. Unlike coordinate-wise quantizers, BlockQuant quantizes blocks on the sphere, preserving the geometry of rotated embeddings more faithfully. We prove that this block-spherical design improves expected distortion performance, including both reconstruction MSE and expected inner-product distortion. Experiments on real embedding data support these theoretical improvements.

Connectomics requires dense segmentation of large-scale electron microscopy (EM) volumes using models trained on small labeled blocks. In practice, a connectomics lab facing a new raw volume has two options: annotate a small block and train on it, or skip annotation and train on the union of available public labeled datasets. We introduce BrainEM, a benchmark that formalizes these options as two standard tracks: the Same-Source Track (Annotate-and-Train) and the Cross-Source Track (Reuse-and-Train). BrainEM brings together 16 EM datasets, including one in-house mouse dataset that we contribute, spanning multiple species, imaging modalities, and resolutions. We further add large-scale test volumes that bring the evaluation closer to realistic practice, and four diverse test targets in the Cross-Source Track. The data, post-processing, and evaluation metrics are fixed across all methods, and the framework is fully model-data decoupled so that new methods can be added with minimal effort. We conduct a systematic evaluation of six representative baselines covering the main technical directions in EM neuron segmentation, making BrainEM the first systematic evaluation at large scale and across diverse cross-source targets. The results differ from prior small-scale evaluations: rankings on the Same-Source Track no longer match those reported on small CREMI splits, and methods diverge sharply on the Cross-Source Track. Both tracks still leave clear room for improvement. We release BrainEM as a public platform to support unified comparison and faster method development in connectomics. All datasets and code are available at https://github.com/kwinderic/BrainEM

This work proposes the Adaptive Federated Hamiltonian Monte Carlo (AFHMC) for Bayesian federated learning (FL) tasks. The prior work on HMC applications for FL achieves $O(\sqrt{d}/\epsilon)$ communication cost under log-convex distributions, already demonstrating its advantage over the Langevin dynamic-based counterpart. Based on a refined analysis, this paper further reveals the benefit of HMC for the FL regime by incorporating a control mechanism for local node shifts. We prove that AFHMC at most requires $O((d/\epsilon^2)^{1/3})$ communication cost up to a logarithmic term, which is significantly better than the existing results. Moreover, the proposed method adapts to client heterogeneity. Under low-heterogeneity settings (i.e., the target sampling precision $\epsilon^2$ is higher than the heterogeneity level), the communication rate of AFHMC can be as good as $O(\log(d/\epsilon^2))$, resembling the communication costs for FL optimization tools. To the best of our knowledge, this provides the first logarithmic communication complexity result for federated sampling. Our code is available at https://anonymous.4open.science/r/Adaptive-Federated-HMC.

Designing DNA libraries is a key challenge from drug design to protein engineering and synthetic biology. Modern generative models offer opportunities to navigate the design space and propose specific sequences predicted to be effective in-silico. Designing deterministic libraries of specific sequences is however limited by the cost of DNA synthesis -- the synthesis barrier. In contrast, high-throughput multiplexed screening can measure the function of billions of biological sequences in parallel. Harnessing this technology requires the design of randomized libraries with specific design constraints to achieve low synthesis costs. In practice, such stochastic libraries are often chosen heuristically, sacrificing control for scale. Is there a way to bridge AI-based in-silico sequence design with high-throughput experimentation? In this work, we introduce Policy Gradients for Library Design (PGLD). PGLD uses a synthesis-aware parametrization of stochastic DNA libraries and optimizes them against a specified objective function. This allows for designing massive, controlled libraries without being limited by synthesis costs. We show how PGLD enables lab-in-the-loop design of multi-round high-throughput experiments, and large-scale in-vitro DNA sampling from generative models. Finally, we use PGLD to design a library of $\sim 10^6$ unique sequences at a cost of $\sim 700$ USD to explore the mutation space of a broadly neutralizing influenza antibody.

Graph-structured data is crucial in various domains like biology and social networks. Comparing graphs, which is a fundamental problem in graph data analysis, is nonetheless highly challenging. Recently, the Gromov-Wasserstein (GW) distance has provided a principled way to compare two graphs. However, computing the GW distance involves solving a complex non-convex optimization problem, making it computationally expensive, especially when the graphs are large. In this work, we propose a neural approximation of the GW distance, called NeuralGW. In NeuralGW, we use a combination of a graph isomorphism network and a transformer to represent the nodes of two graphs as two sets of vectors, treated as two discrete distributions, on which we compute multiple maximum mean discrepancy values given by different kernels. We then use a multilayer perceptron to convert the vector formed by these values into a single value, which is the prediction of the GW distance. Once trained, the model allows for efficient inference, enabling fast structural comparisons between graphs across diverse domains. We also provide a theoretical guarantee for the generalization ability of NeuralGW. Experiments demonstrate the effectiveness and practical applicability of our approach on real-world datasets, in comparison to baselines.

Aligning language models for both helpfulness and safety typically requires complex pipelines---separate reward and cost models, online reinforcement learning, and primal-dual updates. Recent direct preference optimization approaches simplify training but incorporate safety through ad-hoc modifications such as multi-stage procedures or heuristic margin terms, lacking a principled derivation. We show that the likelihood ratio of the optimal safe policy admits a closed-form decomposition that reduces safety alignment to a density ratio matching problem. Minimizing Bregman divergences between the data and model ratios yields Bregman Safety Optimization (BSO), a family of single-stage loss functions, each induced by a convex generator, that provably recover the optimal safe policy. BSO is both general and simple: it requires no auxiliary models, introduces only one hyperparameter beyond standard preference optimization, and recovers existing safety-aware methods as special cases. Experiments across safety alignment benchmarks show that BSO consistently improves the safety--helpfulness trade-off.

Large Language Models (LLMs) have become increasingly capable as tool-using agents, with benchmarks spanning diverse general agentic tasks. Yet rigorous evaluation of scientific tool use remains limited. In chemistry, recent agents can plan syntheses and invoke domain-specific tools, but evaluations often rely on curated demonstrations, expert assessment, or LLM-as-judge scoring rather than exact, judge-free ground truth. We address this gap with chemical procurement cost estimation, a practical task in which an agent must ground chemical identities, retrieve supplier quotes, select valid purchasable packs, normalize quantities, and compute cost from a reaction description. We introduce ChemCost, a benchmark of 1,427 evaluable reactions grounded to a frozen pricing snapshot covering 2,261 chemicals and 230,775 supplier quotes, supporting scalar scoring and stage-level diagnosis of grounding, retrieval, procurement, and arithmetic failures. To evaluate robustness, we further construct controlled noise-injected views that perturb chemical aliases, quantity expressions, missing fields, and input formatting. Experiments with frontier, open-weight, and chemistry-specialized LLM agents show that tool access is necessary but insufficient for solving the task. The strongest agents reach only 50.6% accuracy within 25% relative error on clean inputs and degrade substantially with realistic noise. Stage-level analysis further shows that failures arise from brittle parsing, ineffective evidence integration, invalid pack selection, and non-convergent tool use.

To surface "objective" fact-check notes, X's Community Notes uses Biased Matrix Factorization (BMF) on crowd-sourced ratings to separate note quality from reviewer ideology. We show that for any correctly specified model that decomposes ratings into quality and ideological alignment, this separation requires fixing an ideological anchor: a reference point for what counts as neutral. We give a characterization for the standard implementation: with $L_2$ regularization, BMF anchors quality at a point proportional to the mean reviewer ideology, recovering objective quality if and only if the mean reviewer is neutral. This parallels the Condorcet Jury Theorem: collective judgment recovers truth when the pool is unbiased on average. When that assumption fails, BMF's recovered ranking shifts toward maximizing aggregate reviewer satisfaction, which coincides with social welfare only when the pool is representative. This is not specific to BMF; the rating matrix admits a continuous family of gauge-equivalent decompositions, each implying a different quality ranking, so no mechanism based on ratings alone can recover quality without implicitly choosing such an anchor. What appears to be routine hyperparameter tuning is in fact a normative choice about whose preferences define quality.

Effective machine learning depends not only on how we model data, but also on what data we choose to collect. While large sequence models have revolutionized data modeling, the problem of automated data selection, or "intrinsic curiosity", remains a significant challenge. Classic approaches incentivize exploration by rewarding an agent based on its "learning progress", which measures how much a newly acquired observation improves a world model's predictive ability. However, evaluating these rewards traditionally requires expensive inner loops of gradient descent updates within each trajectory, rendering them computationally impractical at scale. In this work, we investigate whether the emergent in-context learning (ICL) capabilities of sequence models can eliminate this bottleneck by serving as immediate, update-free world models. Specifically, we evaluate whether an exploration policy can be trained to maximize learning progress, using solely the prediction errors and counterfactual context manipulations of an in-context learner. We first prove that in general Markov decision processes, this is in fact impossible in an unbiased way: the resulting intrinsic rewards either suffer from nuisance terms that bias their estimation of true learning progress, or they cannot be implemented using an in-context learner's prediction errors. Conversely, we prove a positive result for a broad subclass of non-temporal settings, encompassing active learning and Bayesian Experimental Design: here, ICL-derived rewards successfully bound and asymptotically converge to the true learning progress. We corroborate our theory with controlled experiments across continuous and symbolic environments, demonstrating that our ICL-driven framework successfully trains curious data-collection policies that explore optimally.

Dynamic Sparse Training (DST) is an effective paradigm for sparse to sparse training of neural network connectivity under a fixed parameter budget. Recent advances in network-science for AI introduced epitopological learning methods, such as Cannistraci-Hebb Training (CHT), demonstrating that network automata applied to the mere network topology enables gradient-free predictions of the sparse connectivity evolution that can trigger significant increase in task performance. However, these methods were currently developed for standard MLP-like fully-connected architectures and need significant rethinking to be extended to convolutional neural networks (CNNs) due to element-wise weight sharing and spatially repeated interactions. In this work, we bridge this gap by introducing epitopological learning DST for CNNs through channel-wise network modeling to address the prohibitive complexity of element-wise modeling, which treats each kernel-channel as a single network node. Results show orders-of-magnitude improvements in time complexity of link-regrowth when applied at channel-level with respect to element-level. We also enhance the channel-wise formulation with lightweight contextual modulation to improve expressivity. Extensive experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet with ResNet and VGG architectures demonstrate that our method achieves performance comparable to dense training using only 30% of the parameters, which significantly reduces computational cost. Furthermore, our approach improves robustness under noisy inputs. These results highlight the effectiveness of topology-aware sparse training and the benefit of decoupling graph construction from sparse update rules in convolutional networks.

Image Manipulation Localization (IML) is commonly formulated as a fully supervised learning task that estimates the optimal manipulation mask $y$ for a given image $x$. In this work, we first reveal the latent nature of artifacts and thus reinterpret IML as a latent-variable problem, $P(y| x)=\int P(y|z) P(z|x)dz$, where $z$ denotes the artifacts. Following this interpretation, we pinpoint the cause for the current IML models' insufficiency as their implicit artifacts modeling strategy, highlighting the necessity of modeling $z$ in an explicit manner. Without direct labels, feature disentanglement is the most appropriate solution for this explicit modeling. Accordingly, we propose a two-stage learning paradigm with the Pairwise Artifacts Learning (PAL) and Standard Localization (SL) phases to estimate $P(z|x)$ and $P (y|z)$ via edit relations. To support our edit-relation-based learning, we further curate EditGroup-45K, a source-anchored dataset organized into edit groups for pair construction. Extensive experiments show that our PAL paradigm yields consistent improvements across diverse IML architectures, and empirical analyses further verify that PAL does capture artifacts explicitly through feature disentanglement. Code and dataset will be publicly available.

Video detectors are usually evaluated on observational splits, where target events co-occur with environmental and capture factors. A high AUC under this setup can reflect either the event itself or its surrounding conditions, and standard reporting cannot tell them apart. This work addresses the entanglement with CARVE, a counterfactual video editing protocol. For each source clip, CARVE generates a matched quartet $(V^0, V^E, V^A, V^{AE})$ that holds camera pose, layout, and surrounding traffic fixed while independently intervening on environment and event. A two-phase generator first produces a layout-conditioned reference image, then performs reference-guided video editing; a three-layer reference / objective / VLM-panel protocol filters the output. Each quartet supports thresholded and continuous diagnostics for false-positive purity, event faithfulness, environment-induced drift, and held-out factor compositions. Thresholded scores are paired with continuous ones so that a conservative detector cannot suppress scores everywhere and look robust. The same audit drives training: CGAA-IC samples weak factors weighted by measured brittleness and adds quartet-level purity, consistency, and faithfulness losses; the inference graph is unchanged. On CCTV accident detection, the audit shows that clean AUC does not predict robustness rankings: a VideoMAE detector with $91.4\%$ AUC on TAD scores only $72.3\%$ CPS on the night slice of CARVE-Q. CGAA-IC raises purity from $84.2\%$ to $89.9\%$ and cross-dataset average AUC from $80.3\%$ to $84.5\%$.

Transfer learning (TL)-based policy adaptation in deep reinforcement learning (DRL) usually relies on target-side data information for retraining when facing new tasks. However, two important issues remain underexplored in practical scenarios: existing DRL transfer learning methods usually lack theoretical guarantees against adversarial attacks, and target-side data information may be unavailable for retraining. In this paper, we propose Certifiable Aggregation of Smoothed Teachers (CAST), which adapts source policies by aggregating multiple certified smoothed teachers without any form of retraining, and further certifies the robustness of the adapted policy at both the action level and the cumulative reward level. CAST certification faces three key challenges: (i) value-function transfer in DRL cannot preserve certified robustness without robust retraining; (ii) reward changes break the direct reusability of source teachers' action-level certificates; and (iii) existing certified robustness transfer results provide limited guidance for certifying the student's cumulative reward. These challenges prevent us from directly using existing TL techniques in DRL. To address them, CAST constructs robustness signals from source certificates, incorporates them into policy aggregation to obtain action-level certificates, and bridges the adapted policy's action-level certificates to its target-task cumulative-reward certificate. Experiments on five multi-objective DRL benchmarks show that CAST exceeds the best source teacher in 17 out of 40 attack configurations and largely remains within the performance range of the source teachers. When testing whether smoothing benefits can be transferred, CAST improves in 39 out of 40 configurations, with a maximum relative gain of 183.9\%.

Recent AI regulations increasingly emphasize the need for mechanisms that preserve the utility of data for AI innovation while preventing misuse, particularly by enforcing purpose limitation in downstream AI applications. In practice, enforcing this principle remains challenging, as released data can be trivially fed into arbitrary models beyond its declared intent. Existing approaches attempt to mitigate this risk by either perturbing data or retraining models to limit unintended use. These strategies, however, offer no protection against inference by unknown or externally trained models, or fundamentally rely on control over the training or deployment. In this work, we introduce non-transferable examples (NTEs), recoded data that act as a task-level "ciphertext" decodable only by a designated model. Where adversarial examples exploit sensitive input directions, NTEs use the complementary insensitive subspace: a training-free, data-agnostic recoding within a model-specific low-sensitivity subspace preserves the authorized model's outputs while degrading unauthorized ones through subspace misalignment. We establish formal bounds certifying output fidelity for the authorized model and showing unauthorized degradation scales with measurable spectral misalignment between models. Empirically, NTEs preserve authorized-model performance across diverse vision backbones and vision-language models, while unauthorized models collapse even under adaptive reconstruction attacks. These results establish NTEs as a practical means to preserve intended data utility while preventing unauthorized exploitation. Our source code and visual demos are available at: https://github.com/model-specific/non-transferable-examples.

Large multimodal models often benefit from commonsense priors, but these priors can mislead reasoning when a task specifies a local mechanism that conflicts with real-world regularities. Existing scientific VQA and visual reasoning benchmarks tend to align problem mechanisms with commonsense, so a correct answer may reflect either mechanism following or prior-based answering. We introduce Causal-ConflictBench, a diagnostic benchmark that makes this ambiguity observable by constructing samples where the current cause-to-effect mechanism contradicts the default commonsense mechanism. CausalConflictBench contains two modules: Textual Rule Override (TRO), which provides explicit commonsense-conflicting textual rules in real science image questions, and Visual Counter-Commonsense Induction (VCI), which requires models to induce a conflicting mechanism from a three-frame visual sequence. Beyond sample-level accuracy, we report Group-Strict Accuracy, factual-prior fallback metrics, and output-level process diagnostics. Across 14 proprietary and open-source multimodal models, we find that high sample-level accuracy can mask unstable mechanism following; errors concentrate strongly on factual-prior answers; and failure modes differ across rule-delivery paths, with TRO revealing rule-application failures and VCI revealing visual rule-induction failures. CausalConflictBench therefore provides a controlled diagnostic setting for uncovering commonsense fallback hidden beneath aggregate accuracy.

Causal discovery methods produce data-driven hypotheses about causal relationships, which may become new scientific findings or inform decisions about downstream interventions. In such settings, false positive causal claims can be more consequential than false negatives. Yet few methods target false positive error control in causal discovery, and existing error guarantees are often asymptotic and require strong versions of the faithfulness assumption. We introduce the $m$-conservative PC algorithm and prove that it controls false discovery rate in the linear Gaussian setting under an extremely weak version of the faithfulness assumption. In our framework, false positive error for causal graphs is defined based on the conditional-independence information they entail, yielding a metric that penalizes both false adjacencies and incorrect edge orientations. The parameter $m$ quantifies a trade-off between algorithmic complexity and the strength of the required faithfulness assumption. The $m$-conservative PC algorithm combines undirected graph learning, a Benjamini-Yekutieli-style procedure for conditional independence testing, and a conservative orientation step that marks unresolved or contradictory edge directions as ambiguous. Our algorithm produces valid outputs and retains error guarantees even in the finite-sample setting, despite possible random errors in the conditional independence tests.

Latent confounding poses a significant obstacle to identifying the causal effect of a treatment on an outcome. To address this, many existing studies leverage proxies of the latent confounder to indirectly adjust for the confounding bias. However, they impose specific structural constraints on the proxies and typically require multiple proxies. In this paper, focusing on the latent variable linear non-Gaussian acyclic model (lvLiNGAM), we propose a causal effect identification procedure requiring only a single agnostic proxy. Crucially, the term "agnostic" means that the causal connections between the proxy and the treatment-outcome pair can be arbitrary and are not required to be known a priori. This structural complexity precludes identifying the causal effect via a simple closed-form formula. Consequently, our identification procedure is designed to first derive candidate solutions from cumulants and then isolate the valid solution by examining certain independence relationships. We present a series of new theoretical results, which collectively establish the soundness of our identification procedure: given the observational population distribution, it correctly identifies the true causal effect when identifiable, and correctly reports unidentifiability otherwise. Finally, we empirically validate our theoretical results.

In Large Language Model (LLM) training, data mixing plays a pivotal role in determining model performance. Recent methods optimize mixture weights via proxy models, but they rely on the assumption of static data distributions. As a result, when the underlying data pool shifts, these methods require costly retraining from scratch. This limitation restricts their ability to scale seamlessly from small settings to larger data pools and model sizes. In this paper, we propose CausalMix to address this limitation by casting data mixture optimization as a causal inference problem. We formulate the statistical features of the data pool as covariates and the domain mixture as the treatment. After fitting a causal model on 512 runs of Qwen2.5-0.5B to estimate the Conditional Average Treatment Effect (CATE), we extrapolate the optimal mixture for an 800K data pool and apply it to train a 7B model. Furthermore, we successfully generalize the framework to long chain-of-thought data on Qwen3-4B-Base. By leveraging causal modeling to isolate confounding biases, CausalMix dynamically infers state-dependent optimal data mixtures. Extensive experiments show that the mixture guided by CausalMix consistently improves performance across multiple downstream tasks, outperforming RegMix and other baselines. In addition, we use the CATE Interpreter to provide visual analysis of the learned mixing strategy. Overall, CausalMix offers a causal and interpretable framework for optimizing LLM data mixtures.

Language model agents are becoming proficient executors at isolated, short-horizon tasks such as software engineering and customer service. Yet real-world challenges require a combination of sophisticated skills that remain largely untested in agents: (1) navigating long horizons amid uncertainty; (2) acquiring information in noisy environments; (3) adapting to a changing world; (4) orchestrating multiple moving parts toward a coherent goal. We introduce CEO-Bench, which evaluates these capabilities together by simulating a representative real-world task: operating a startup for 500 days. Given diverse and realistic company management tools, business databases, and social media, an agent needs to design pricing strategies, allocate operating budgets, analyze business data, respond to unexpected competitor moves, and more. Our evaluation shows that most state-of-the-art models struggle to succeed in this environment, and only one model (GPT-5.5) finishes the simulation above its $1M starting balance. CEO-Bench reimagines the role of AI in the future, shifting from solving isolated tasks to driving sustained, adaptive progress over time. We open source full code and trajectory.

We investigate the use of formal methods to provide tight and sound generalization bounds for learning algorithms. By casting the traditional notion of algorithmic stability as a specification to be verified, we demonstrate that recent advances in reachability analysis can yield provable bounds on the generalization of a given model and algorithm on a sample dataset. As sample-specific algorithmic stability is insufficient to bound the usual distributional notion of generalization, we develop a novel concentration inequality to connect the sample-specific results of formal certification algorithms to the required distributional analysis.The resulting framework enables the theoretical analysis of prior generalization bounds to extend far beyond their original restrictive analytical assumptions while achieving a provably sound bound on the generalization gap. In practice, we demonstrate that our framework provide formal generalization guarantees that are orders of magnitude tighter than alternative computational approaches at scales ranging from toy datasets to fine-tuning of modern large language models. While we implement certification-enhanced versions of several well-known stability results, future extensions of our approach will enable tighter bounds and enhanced practical adoption across the spectrum of modern generalization bounds.

Imitation learning policies for robotic manipulation inherently suffer from compounding errors. To mitigate this, inference-time policy steering introduces predictive reasoning to evaluate and refine action proposals before execution. However, existing methods typically anticipate consequences in unstructured pixel or latent spaces, leading to physically inconsistent predictions and unreliable error detection. Furthermore, their blindness to global progress often yields locally plausible corrections that fail to complete the overall task. To address these limitations, we propose Chain-of-Correction (CoC), an inference-time steering framework driven by a single VLM. Shifting away from unstructured state predictions, CoC explicitly leverages anchor-grounded scene graphs to track exact 3D physical relations. Built upon this representation, a progress-aware dual-check mechanism verifies historical execution and predicts future task advancement to systematically intercept myopic action proposals. Through a failure-augmented supervised fine-tuning pipeline, CoC seamlessly unifies scene graph extraction, progress evaluation, and precise kinematic correction, eliminating the need for task-specific world models. Extensive experiments across the COLOSSEUM benchmark, a newly introduced offline protocol for predictive reasoning, and real-world tasks demonstrate that CoC significantly outperforms state-of-the-art baselines in fine-grained error detection and robust error recovery.

Long-term semantic navigation requires a robot to reuse past observations after appearance and scene change, but semantic memories are only useful if the robot can relocalize into the memory without corrupting it with false visual matches. We propose CROSS, a change-robust topological memory that introduces a pre-commitment localization layer between visual place recognition and map update. Instead of treating a retrieved keyframe as an immediate place association or loop-closure factor, CROSS lifts each RGB-D retrieval into a candidate global $\mathrm{SE}(3)$ pose mode using relative pose estimation. A bounded Gaussian-mixture filter then propagates competing continuous trajectory branches with odometry, rejects branches that are physically inconsistent, and promotes only persistent branches to loop closures. This moves ambiguity handling from discrete place IDs or post-hoc graph-factor rejection to continuous pose-space validation before map commitment. Across public long-term relocalization benchmarks and real quadruped object-navigation experiments, CROSS improves reuse of a single sparse RGB-D memory under illumination, seasonal, dynamic-scene, and object-level change.

We present Channel-wise Vector Quantization (CVQ), a novel image tokenization paradigm that replaces patch-wise tokens with channel-wise tokens. Unlike conventional vector quantization, which assigns a discrete token to each patch feature vector, CVQ quantizes each channel of the feature map. This formulation represents an image as discrete levels of visual details, rather than as a grid of spatial patches. Based on CVQ, we introduce a new visual autoregressive framework with "next-channel prediction". Instead of rendering images patch by patch in raster order, our Channel-wise Autoregressive (CAR) model predicts image channels sequentially, producing progressively enriched visual details. Specifically, it first sketches global structure and then refines fine-grained attributes, akin to a human artist's workflow. Empirically, we show that: (1) CVQ achieves 100% codebook utilization with a 16K+ codebook size without any bells and whistles, while reducing reconstruction FID by 50% over conventional VQ; and (2) CAR outperforms the AR baseline by improving the GenEval score from 0.69 to 0.74 and the DPG score from 79.86 to 82.14, demonstrating strong effectiveness for text-to-image generation. We hope our research offers a new perspective on the fundamental unit of visual tokenization by moving from spatial patches to channels.

This paper aims at analyzing the regularization effect that data augmentation induces on supervised regression methods in the proportional regime, where the number of covariates grows proportionally to the number of samples. We provide a tight characterization of the test error, measured in mean squared error, in terms only of the population quantities of the true data, as well as first and second order statistics of the augmentation scheme. Our results are valid under misspecified feature maps, and for any network architecture where only the last readout layer is trained, and the rest of the network is either frozen or randomly initialized. We specify our results in the case of Gaussian data, and show that our asymptotic characterization is tight in this setting.

LLMs are powerful but incur substantial inference cost, which can be exploited by malicious users to induce overly long outputs, increasing latency and operational expense. Prior resource consumption attacks either coerce repetition or suppress the end-of-sequence token by lowering its logit value. However, with autoregressive, samplingbased generation, the search space grows exponentially with length and small deviations or unstable trajectories can derail optimization, limiting attack effectiveness and stability. To address this, we introduce Chatter Attack, a novel class of resource consumption attacks that optimize adversarial prompts to induce predefined short target phrases, thereby avoiding search-space explosion while still eliciting long responses. Specifically, Chatter Attack triggers self-doubt using short doubt-inducing phrases as the target, and sustains this state throughout generation via inference attention loss and global entropy loss. To extend to black-box settings, we further propose a novel Bayesian Prompt Optimization (BPO) method, which models the objective function globally by constructing a discrete token kernel, efficiently exploring the entire search space using prior information. Extensive experiments across multiple models and datasets show that our method outperforms baselines, with maximum response length increases to 31.5×, and revealing practical resource-depletion risks for LLM services.

World models require state tracking, which is the ability to maintain a correct latent state across action sequences. Existing benchmarks are often synthetic or language-based, limiting their value as tests of structured state updates in realistic domains. We introduce Chess-World-Model, a large-scale state-tracking benchmark built from $10$ million real chess games, where models predict the exact board state reached after a sequence of legal moves. Alongside a held-out real-game split, we include an out-of-distribution split from uniformly random legal play, which tests whether models learn the transition rules rather than shortcuts from common human positions. Prior theoretical and empirical work has shown that Transformers struggle to state-track, while input-dependent linear RNNs require expressive state-transition matrices to do so. We therefore benchmark a causal Transformer, block-diagonal SLiCE, Mamba-3, and Gated DeltaNet with negative eigenvalues under a matched interface and training protocol. The recurrent models strongly outperform the Transformer at $3$ and $8$ million parameters. Real-game performance saturates above $18$ million parameters, but the random-uniform split remains discriminative up to $38$ million, exposing failures otherwise hidden by scale. Additionally, ablations show that less expressive state-transition mechanisms reduce performance on the out-of-distribution split for all three recurrent models. Together, these results establish Chess-World-Model as a practical large-scale benchmark for state tracking that exposes failures model scale would otherwise conceal.

Large language models (LLMs) rely on long-horizon tool invocation sequences for complex tasks, where each invocation can alter the task state and condition subsequent decisions. In long-horizon tool use, final-outcome rewards provide weak credit assignment over long interaction traces. Step-level rewards can offer more targeted feedback, but obtaining reliable step supervision often requires human or LLM judgment, or additional rollouts to estimate the downstream effect of an intermediate decision. In this paper, we argue that effective tool-use agents should estimate the long-horizon value of a possible next tool invocation before executing it. This objective requires comparative supervision over alternative invocations under the same context, while logged trajectories only contain the invocation that was actually taken. Therefore, we propose Comparative Inference for Tool-use Agents (CITA). CITA trains a Comparative Inference Model (CIM) from paired signals that combine observed tool behavior, scalable supervision from a Bayesian tool-graph simulator, and semantic judgments from LLM-based comparison. The resulting CIM learns to estimate how likely a possible next tool invocation is to support final task success under the current context. Across three tool-use benchmarks and multiple backbone LLMs, CITA consistently improves Tool F1 and task success. Additional analysis shows that CIM learns accurate step-level value estimates for comparative tool choices.

Cinematic understanding requires more than recognizing visible events: models must acquire dialogue evidence, bind it to speakers, and use it to infer narrative meaning. Existing video benchmarks evaluate pieces of this problem, including video question answering, audio-visual reasoning, and temporal grounding, but rarely test whether these abilities form a coherent cinematic evidence chain within the same clips. As a result, a model may answer a movie question correctly while failing to localize the relevant utterance, identify the speaker, or integrate speech with visual context. We introduce **CineMME**, a human-verified benchmark designed to diagnose speech-grounded cinematic reasoning, supported by approximately $1,600$ expert person-hours of annotation. CineMME contains $975$ curated movie clips and two linked tracks: *CineMME-Grounding*, with $13,554$ timestamped speech segments and $33,083$ actor face boxes for temporal dialogue localization, transcription, and active-speaker grounding; and *CineMME-Reasoning*, with $1,600$ multiple-choice and $316$ open-ended questions spanning perception, audio, narrative, social, and cinematic understanding. We evaluate $20$ MLLMs on reasoning and $10$ omni-modal models on grounding. The results reveal a pronounced who-what-when decoupling: Gemini-3 Pro localizes dialogue in time 68.33% tIoU) and partially transcribes it (WER 29.00%), yet reaches only 13.24% sIoU for active-speaker grounding while achieving 65.88% reasoning accuracy. Modality ablations further show that vision is not uniformly beneficial: it improves perception-heavy questions but can hurt dialogue-centric reasoning. CineMME provides a compact diagnostic testbed for measuring whether multimodal video models truly connect dialogue, speakers, visual evidence, and narrative meaning, rather than succeeding through disconnected cues.

We introduce CIPHERGRID, a rule-inference and path-finding benchmark. In each task, models are given example grid-world images paired with encoded descriptions and encoded solutions, then receive a new encoded grid-world to solve. To answer correctly, they must infer which symbols denote tiles, actions, row boundaries, and game mechanics, decode the new world, and produce an encoded path from the start cell to the target cell. The benchmark is designed so that each component is individually solvable: rule inference is calibrated against human performance, and path-finding is solver-verifiable via dynamic programming. Although the underlying skills are individually solvable and verifiable, CIPHERGRID tests their interaction: whether models can infer an unfamiliar symbolic system, and carry it through rule application, and planning. CIPHERGRID remains difficult for current frontier models, which only achieve between 18.74\% and 42.20\% accuracy. CIPHERGRID offers a controlled test of whether models can coordinate multimodal grounding, rule inference, and sequential planning under unknown symbolic relationships.

Traditional backdoor attacks inject artificial trigger patterns into training data, causing models to misclassify triggered inputs while maintaining normal behavior on clean samples. Since these artificial triggers are absent in clean datasets, standard knowledge distillation (KD) typically eliminates them, leading to the common belief that KD serves as an effective purification defense. In this paper, we propose DAR (Discover-and-Relabel), a support-persistent backdoor method that retains its behavior even after clean-data distillation. DAR identifies naturally abundant patterns in low-dimensional feature subspaces using subspace clustering and applies label-only poisoning to samples that inherently satisfy the discovered rule. At test time, the backdoor is activated by locally modifying the corresponding subspace of the target images. By instantiating spatial and frequency operators, DAR achieves attack accuracy comparable to state-of-the-art backdoor attacks in ImageNet classification and CLIP-based prompt tuning. Notably, DAR keeps attack success above 96% after clean KD and remains defense-resistant in practice.

Deep models for medical time series can achieve strong benchmark accuracy yet remain brittle under acquisition artifacts, physiological variability, and low-resource training conditions. We present \textbf{ClinStab}, a Medformer-based framework for EEG and ECG classification that studies whether channel-aware dual-stream interaction and perturbation-aware optimization can improve the clean/robustness tradeoff. The proposed \textbf{S}alience-\textbf{C}ontext \textbf{D}ual \textbf{M}odulation (SCDM) module uses energy as a routing prior: high-energy channel-aligned components receive attentive modeling, while residual components remain trainable through a lightweight context pathway. ClinStab then combines clean supervision, perturbed supervision, prediction consistency, and teacher-guided stabilization under adversarial perturbations. Across four public datasets and 11 baselines, ClinStab improves accuracy over Medformer on all datasets and improves selected robustness metrics under the perturbation settings studied here, while exposing tradeoffs on PTB ranking metrics and soft-routing alternatives. These results support ClinStab as a practical same-backbone stability intervention, not as evidence of universal clinical robustness.

Machine unlearning for large language models optimizes two opposing losses: a forget objective drives the model away from targeted knowledge, while a retain objective holds it near its original behavior. Existing methods commit to a fixed schedule of learning rate, retain weight, and step count chosen by offline grid search, which is blind to how the trajectory actually evolves. We show that LLM unlearning trajectories share a method-agnostic three-phase structure, an early phase where the two gradients are nearly orthogonal, a conflict phase where they oppose each other and retain loss begins to grow, and a terminal phase where representational divergence crosses an irreversibility threshold. Each phase calls for a different control policy. We propose CLUE, a closed-loop framework whose architectural choices each correspond to a phase or transition: a dual-tower controller separates forget and retain signal pathways, a conflict-driven gate switches between them, and an explicit collapse-risk head supervises anticipatory stopping. The controller is trained by task-distribution meta-optimization. We provide a single-step bound on conflict-induced retain loss growth, a quantitative irreversibility result, and a stopping regret bound. On TOFU, WMDP, and MUSE with Llama-2/3 and Mistral models from 7B to 70B, CLUE achieves better forget-quality versus model-utility trade-offs than fixed-strategy baselines, stops within 5\% of the oracle, and remains robust under adversarial prompting, relearning, and INT4 quantization.

Joint classification of hyperspectral imagery (HSI) and Light Detection and Ranging (LiDAR) data benefits from complementary spectral--geometric cues, but existing fusion methods largely assume locally consistent cross-modal features and symmetric pixel-wise trust, which is often violated in real urban scenes. We term this pixel-level phenomenon \emph{cross-modal local inconsistency} (CMI) and quantify it with a Strong-Edge CMI Ratio: dual-modal gradient analysis shows that over $16\%$ of strong-edge pixels exhibit single-modality dominance across three benchmarks, while ROI-level analysis reveals the class-entanglement induced by symmetric fusion. To address this issue, we propose \textbf{CMI-Trans} (\textbf{C}ross-\textbf{M}odal \textbf{I}nconsistency-aware \textbf{Trans}port), which promotes per-pixel reliability modeling from a post-hoc diagnostic to a structural signal for alignment and training. CMI-Trans combines Energy-Calibrated Reliability (ECR), Uncertainty-Shaped Transport Alignment (USTA) with Sinkhorn-regularized optimal transport, and Triadic Co-Optimization (TCO) that jointly optimizes classification, uncertainty consistency, and alignment stability on a Mamba-based fusion backbone. On Houston, Trento, and MUUFL, CMI-Trans achieves competitive overall accuracy and substantial gains on CMI-affected classes (e.g., $+17.90\%$ on Houston C10). The ECR-derived uncertainty score also separates reliable from erroneous predictions ($\rho(U,e)<0$; top-$20\%$ highest-$U$ error rate $\le 0.18\%$), while improving feature separability in shadow-induced CMI regions. These results show that explicitly modeling CMI through uncertainty-shaped adaptive alignment yields more accurate and interpretable multimodal remote sensing classification.

Vision-language-action (VLA) models have emerged as a promising paradigm for autonomous driving trajectory planning. While explicit test-time thinking has driven mainstream progress in Large Language Models (LLMs), how to effectively utilize reasoning in driving VLA remains unsettled. We revisit this gap through Coarse-to-Refine (C2R), a trajectory self-refinement framework for driving VLA that completes refinement in a single autoregressive pass. C2R first predicts a coarse trajectory, performs explicit reasoning over it, and then generates a refined trajectory autoregressively. We show that the refinement formulation itself drives imitation learning improvement while enabling reinforcement learning to utilize reasoning content for further optimization. C2R establishes VLA state-of-the-art performance across standard and challenging benchmarks, achieving 92.1 PDMS on NAVSIM v1, 90.4 EPDMS on NAVSIM v2, and 49.6 EPDMS on NavHard.

We formalize and enable the task of open tree decomposition, which segments an image into hierarchical trees of visual components with unconstrained granularity and flexibility. Specifically, we provide the foundation benchmark for this new paradigm with the following three key contributions. First, we overcome the prohibitively high cognitive and physical bottlenecks of manual annotation by developing a fully automated generation pipeline that synergizes the semantic reasoning of Large Vision-Language Models (LVLMs) with the precise geometric grounding of SAM 3. Second, leveraging this pipeline, we construct COCOTree, a massive-scale benchmark featuring over 21K images and 1.8M structural nodes. By embracing an open-vocabulary space of over 3.5K unique labels, it successfully captures the long-tail distribution of complex physical assemblies. Notably, rigorous human evaluation confirms our generated annotations demonstrate strong alignment with human structural judgment. Third, we establish a standardized evaluation protocol by proposing the Open Tree Quality (OTQ) metric, which jointly assesses mask precision, label accuracy, and structural consistency. We release our dataset and benchmark code at https://github.com/melonkick3090/COCOTree.

Change Detection & Captioning (CDC) aims to jointly localize changed regions and describe their semantic content from bi-temporal remote sensing images. While recent approaches have demonstrated the effectiveness of unified modeling, they primarily rely on shared representations or jointly optimized architectures, leaving task-specific communication under heterogeneous supervision less explored. However, detection and captioning differ fundamentally in their supervision granularity, leading to heterogeneous representations with uneven reliability across tasks and spatial locations. In this paper, we reformulate CDC as a selective information routing problem between heterogeneous task branches, where the key challenge is to determine when and where information should be transferred. To this end, we propose CoDeRNet, a unified framework that preserves task-specific representations while enabling bidirectional exchange of complementary information. CoDeRNet introduces a Confidence-Decomposed Routing (CoDeR) mechanism, which decomposes the routing condition into receiver-side need for support and sender-side reliability cues to guide selective transfer. Experiments on LEVIR-MCI and WHU-CDC show that CoDeRNet achieves strong overall CDC performance, particularly improving caption generation while preserving competitive change localization. Extensive analyses further show that the benefits arise not merely from joint learning or generic cross-task communication, but from explicitly modeling selective cross-task information routing. These results highlight the importance of conditional information transfer for unified CDC under heterogeneous supervision.

Claims about the cognitive capabilities of large language models are typically tested by translating behavioral paradigms into natural language prompts, bypassing the visual interfaces and time-pressured interactions through which human participants actually engage with experiments. In parallel, recent demonstrations that AI agents can produce human-like response data on live behavioral tasks raise concerns that online datasets used across the social and behavioral sciences are increasingly susceptible to contamination. We introduce CogArena, a benchmark of interactive experimental paradigms drawn from the social and behavioral sciences. Agents interact with the experiments through a standard browser, processing visual stimuli and responding under configurable deadlines. We designed a three-level scoring pipeline that evaluates task completion, performance accuracy normalized against literature-derived human baselines, and alignment with canonical task-specific signatures. Across frontier multimodal agents, we find that agents complete the tasks, but their behavior diverges from humans on the signatures these paradigms reliably elicit in people. CogArena provides the tasks, API, scoring pipeline, and a public leaderboard for ongoing community submissions.

We introduce CoLa3D, a simple and effective model for composable 3D scene decomposition that, given a single whole-scene latent from either a single image or an existing scene mesh as input, directly outputs a set of object meshes with consistent spatial layout that can be assembled back into a coherent scene. CoLa3D introduces a novel paradigm for 3D scene decomposition within a shared \emph{whole}-scene latent, unlike alternatives that perform per-object generation in canonical spaces and then glue the objects back together with fragile pose and scale estimation. The network is a lightweight, flexible, and scalable query-based transformer that accepts various types of queries, including 2D masks, 3D points, and learnable queries, and learns to decompose the scene into objects in a compact latent space. On both synthetic and real-world datasets, CoLa3D achieves state-of-the-art performance on composable 3D scene decomposition, substantially improving global scene-level consistency and object-level quality while running $2\times$ faster and using less than 1% of SAM 3D's training data. Visible depth on par with MoGe2 further confirms that latent decomposition preserves accurate visible scene geometry while producing physically complete, decomposable object meshes.

Achieving high-fidelity object-level control in diffusion transformers remains a significant challenge despite the introduction of structural priors like depth and Canny maps. Current object-level conditional generation methods frequently suffer from visual artifacts and struggle to maintain precise control over objects within small localized regions. To address these limitations, we propose Cascaded Object-Level Latent Refinement (COLLAR), a training-free framework that progressively optimizes object-level features via the Field-of-View (FoV) expansion. First, we propose the Cross-Scale Semantic Alignment (CSSA) module to address spatial-semantic gaps by injecting object-level features into extended-FoV branches via attention mechanisms. To further optimize these features, the Cyclic Feature Injection (CFI) module introduces a reciprocal background feedback mechanism. It leverages a frequency-based adaptive strategy to selectively update the global backbone with context-aligned local information. Finally, the extended-FoV branch serves as a hub for feature optimization, ensuring that object-level features are integrated into the global generation process without compromising final image quality. Extensive experiments on the COCO-MIG and COCO-POS benchmarks demonstrate that our approach consistently outperforms state-of-the-art methods across semantic alignment, image quality, and spatial fidelity.

Modeling biomolecular conformation and dynamics is critical for elucidating biological functions, motivating generative surrogates for molecular dynamics (MD) simulations to address two complementary tasks: time-independent conformational sampling and time-dependent trajectory generation. Fundamentally, both tasks require matching the distribution of the generated collections to the MD reference, rather than reproducing individual configurations. Yet existing deep-learning approaches predominantly optimize a per-sample regression loss (score- or flow-matching), whose gradient is computed in isolation per sample and therefore carries no direct signal about the cross-sample distributional properties that define both tasks, leading to biased ensembles and long-horizon temporal drift. We address this with Collective Supervision, a training paradigm that aligns the empirical measures of the generated and reference collections via a maximum mean discrepancy on SE(3)-invariant physical observables; a single loss handles both tasks within a unified formulation. To mitigate the exposure bias inherent in autoregressive trajectory rollout, we further introduce Collective Rolling Forcing, which couples Collective Supervision with autoregressive self-rollout during training. Our framework, CoDyna, generalizes across proteins, protein--ligand and protein--protein complexes; on four all-atom MD benchmarks (ATLAS, MISATO, DynaRepo, and DynaBench), a unified model surpasses task-specialized baselines on most thermodynamic and kinetic fidelity metrics and remains structurally stable over 2k-frame rollouts.

Text-to-image (T2I) models have advanced considerably in generating high-quality images from textual descriptions. However, their ability to associate colors with concepts remains largely constrained to explicit color names or codes, while their capacity to handle implicit concepts, such as emotions and visual states, remains underexplored. To address this gap, we introduce ColorConceptBench, an expert-annotated benchmark that systematically evaluates color-concept associations through probabilistic color distributions. ColorConceptBench moves beyond explicit color specifications by examining how models interpret 1,281 implicit color concepts, grounded in 6,584 human annotations. Our evaluation of nine leading T2I models reveals that performance varies substantially across semantic categories, and models exhibit a significant lack of sensitivity to abstract semantics. These limitations persist even when applying classifier-free guidance scaling at inference time, suggesting that achieving human-like color understanding demands a shift in how models learn and represent implicit semantic meaning.

Vision--language models (VLMs) such as CLIP retain zero-shot recognition ability, but their open-world use often requires task-agnostic continual learning: tasks arrive sequentially, past data are unavailable, task identity is unknown, and the backbone cannot be retrained. We argue that this setting fails VLMs through three coupled pressures beyond stability--plasticity tradeoff: (1) modal-consistency drift, (2) representation-space interference, and (3) per-task capacity shortfall. We propose COMET (COllaborative Mixture-of-Expert Transfer) to resolve these three issues respectively as follows. A frozen teacher provides only a feature reference on a shared image--text pool; Tri-Affinity Distillation preserves image--image, text--text, and image--text geometry; and Student-Aware Cross-Expert Distillation aligns new experts with compatible prior experts so routing errors degrade gracefully. We design COMET to keep capacity and inference separate: a contextual bandit keeps or merges full per-task experts using student-side signals only, while a reconstruction router selects sparse experts at test time without teacher semantics. On the X-TAIL benchmark, we show that each proposed components indeed improves the continual learning performance. COMET shows that continual vision--language adaptation can preserve multimodal geometry, coordinate accumulated experts, and allocate capacity jointly when feature reference, expert shaping, routing, and capacity control are separated by construction.

Decentralized training enables large-model training over low-end GPUs and internet-grade connections, but communication along both data-parallel and pipeline-parallel axes becomes the primary bottleneck. We study post-pretraining adaptation in this setting. We propose an asynchronous two-circuit system: a fast compressed training circuit drives throughput using activation masking for pipeline-parallel (PP) transfer and compressed data-parallel (DP) synchronization, while a slow anchor circuit runs occasional unmasked forward--backward passes off the critical path. Then, we introduce a spectral correction optimizer that uses these delayed anchor priors to denoise masked gradients without blocking the fast stream. Although prior work has found aggressive activation compression unreliable, we show that masking supports post-pretraining adaptation at high compression rates when anchored this way. Pipeline-parallel compression alone yields up to a 9x throughput gain, and combining it with data-parallel compression increases beyond 40x over internet-grade 200Mbps connections, while matching dense uncompressed performance across domain adaptation and continual pretraining.

Reconstructing 3D scenes from sparse views without per-scene optimization remains highly challenging, especially for recovering accurate geometry and fine textures. While recent feedforward approaches leverage generalizable 3D Gaussian Splatting (3DGS) for scene generation, they typically assign one or multiple Gaussians to each pixel. Such uniform allocation not only produces highly redundant representations by wasting primitives in homogeneous areas, but also fails to exploit the inherent geometric priors of planar surfaces, often resulting in sub-optimal structural modeling. To address these limitations, we present CompactSplat , a novel feedforward framework that enables a content-aware allocation of 3D Gaussians. By integrating texture-guided spatial partitioning with hierarchical sampling, CompactSplat adaptively distributes primitives — concentrating dense Gaussians in complex regions while tiling flat areas with expanded, scale-modulated primitives. Moreover, our framework inherently supports dynamic sparsity control at inference time, enabling seamless trade-offs between rendering fidelity and memory footprint without requiring network retraining. Extensive experiments on RealEstate10K, DL3DV, and ScanNet demonstrate that CompactSplat consistently outperforms prior methods on both standard metrics and high-resolution rendering consistency, achieving high-fidelity reconstructions with significantly fewer primitives.

In this work, we provide a complete characterization of the uniqueness of linear patterns for generic fully connected neural networks activated by ReLU or leaky ReLU functions. With this result, together with several tools from geometry and probability, we compare the numbers of linear regions for a wide variety of networks from both theoretical and practical perspectives, with minimal prerequisites. More specifically, on the theoretical side, we show that networks with neurons arranged as $pq\times2$ (2 hidden layers, each having $pq$ neurons) are expected to have more linear regions than those arranged as $p\times 2q$ for $p\ge 1$ and $q\ge 2$; on the empirical side, we show the validity of a Monte-Carlo approach for comparing linear regions via linear patterns, and carry out experiments for networks in various scenarios. Our results indicate that neurons closer to the input layer tend to have more impact on the number of linear regions. Moreover, along the way, we find the precise value of the expected number of linear regions for networks with 2 hidden layers, and an explicit upper bound for networks whose layer widths are non-increasing. The implementations for this paper are available through the following link: https://github.com/temp-repo-259848/Regions-ReLU-Network

Large language models offer a promising technique for translating natural-language optimization problems into solver code, but their reliability remains constrained by the quality of template retrieval. While recent structured methods improve over flat prompting through hierarchical taxonomy search, their traversal rules remain fixed or LLM-driven at inference time, leaving retrieval unable to improve from past successes and failures. This motivates treating template retrieval as a policy that can be learned from modeling experience, rather than as a static inference-time rule. We propose COMPASS, a policy-amortized framework for retrieving optimization templates in LLM-based solver modeling, which represents the template library as a directed acyclic graph and formulates retrieval as a sequential decision process over template nodes. A deep Q-network learns a reusable traversal policy from template-level judgments and execution feedback, reducing reliance on fixed heuristic traversal through feedback-driven graph navigation. To support extensible retrieval, COMPASS embeds problem descriptions and template nodes in a shared semantic space, allowing newly inserted templates to be scored from their text descriptions without redefining the action space. Experiments across multiple LLM backbones and optimization-modeling benchmarks demonstrate that COMPASS consistently improves over heuristic structured retrieval, with the largest gains on complex and mixed-category problems.

Subject-driven generation (SDG) evaluation using multi-modal LLMs (MLLMs) provides better interpretability and stronger alignment with human judgment than traditional embedding-based methods. However, MLLM-as-a-Judge scores are limited in differentiating between error and error-free SDG images. In this paper, we propose CompJudge, which overcomes these limitations by introducing comparative capabilities to MLLM-as-a-Judge and recalibrating scores based on comparison results. To rigorously evaluate the correctness of SDG evaluators, we introduce CompILIAS, a diagnostic benchmark of triplets (reference, identity-preserving image, identity-degraded image) with verified binary ground-truth labels indicating whether the SDG image preserves the reference identity. We demonstrate improvements of SDG evaluation over existing approaches: 8.9pp on CompILIAS and 10.62--15.98pp on the DreamBench++ dataset. These improvements consistently generalize across diverse MLLM backbones and across subject categories.

Aligning human-interpretable concepts with the internal representations of modern machine learning systems remains a central challenge for interpretable AI. We introduce a geometric framework to compare supervised human concepts with unsupervised representations derived from foundation models. We formalise concept frustration: a mismatch that arises when an unobserved concept induces relationships between known concepts that cannot be made consistent within an existing ontology. We develop task-aligned similarity measures that detect this phenomenon, and show that frustration is identifiable in task-aligned geometry while conventional Euclidean comparisons fail. Under a linear–Gaussian generative model, we derive a closed-form expression for Bayes-optimal concept-based classifier accuracy, decomposing predictive signal into known and unknown contributions and identifying where frustration impacts performance. Experiments on synthetic data and real language and vision tasks demonstrate that frustration is present in foundation model representations and that incorporating missing concepts reorganises learned representations to better align human and machine reasoning. These results provide a principled framework for diagnosing incomplete concept ontologies and improving alignment in interpretable AI systems.

Concise reasoning in large language models (LLMs) seeks to generate only essential steps needed to arrive at a final answer, thereby alleviating issues of overthinking. Most proposed approaches scalarize length and reward into a single objective, requiring coefficients or thresholds that must be re-tuned across domains and model scales. We address this brittleness by treating concise reasoning as a constrained problem that minimizes length subject to an accuracy floor, and deriving a tractable algorithm, Performance-Aware Length Update (PALU), which replaces each intractable update of Lagrangian optimization with a tractable surrogate while preserving its structural prescription. On DeepSeek-R1-Distill-Qwen-1.5B, PALU cuts generation length by 64\% and improves accuracy from 44.2\% to 51.6\% across six benchmarks, a level matched by GRPO only with 2.5× more tokens. The Lagrangian dynamics yields a two-phase compression: an initial phase where the residual signal still allows compression without performance cost, followed by a trade-off phase where performance gates further length reduction. The same hyperparameters transfer across math, logic and STEM, and across 1.5B to 14B models, suggesting that constrained optimization is a productive design scaffold for training reasoning LLMs.

Robot manipulation often depends on short-term cues—such as brief contacts, temporary occlusions, and intermediate object states—that are ambiguous in a single frame and can only be resolved by integrating observations over time. However, injecting historical information solely by enlarging the history window is computationally costly and frequently yields limited gains due to substantial redundancy and task-agnostic, inflexible sampling of observations. We propose CondenseVLA, a multi-frame VLA policy with a lightweight learnable history condensation module for using short-term visual context in action decoding. Given a sequence of cached per-frame features from the VLM backbone, CondenseVLA distills observation history into a fixed set of slot tokens using learnable queries and multi-head cross-attention. Simple auxiliary regularization encourages diverse attention patterns by penalizing overlap among slot attention maps and reduces slot 12 redundancy by decorrelating slot embeddings, stabilizing learned condensation without imposing strong time priors. By expanding the history window and distilling it into a compact token set, CondenseVLA improves over the baseline on the evaluated Simpler-WidowX tasks, increasing success rate from 58.4 to 70.8 (+12.4), achieves 97.2% on LIBERO, and reaches 52.0% on real-world tasks. Beyond the gains, our results show that long history windows contain substantial redundancy, so naively providing more frames yields limited benefits, whereas learned condensation into fixed slot tokens provides an effective bounded-token way to leverage temporal context.

Aggregate accuracy hides where models succeed and fail. Estimating conditional performance profiles from gold labels alone is expensive, while cheap auxiliary signals such as LLM-judge scores, pairwise comparisons, confidence scores, and judge-disagreement features can be collected for every benchmark item but are often biased or miscalibrated. We propose \method{} (Local Augmented Control-Variate Evaluation), a semi-supervised estimator for conditional LLM evaluation. The key step is local centering: after subtracting the conditional mean of a cheap signal within the target profile region, any linear augmentation has zero conditional mean and therefore cannot change the estimand. The augmentation coefficient is used only for efficiency, and a local ridge control variate combines a gold-label residual mean from the labeled subset with a cheap-signal mean from the full item pool. We prove calibration-free identification, unbiasedness for grouped profiles, local oracle optimality within centered linear augmentations, and first-order adaptivity to the estimated coefficient. The resulting gain formula depends on a local $R^2$ that can be estimated from the available labeled subset and cheap signals, diagnosing where cheap signals are useful. We extend the same principle to direct paired model gaps and deployment-weighted scores, and validate it on MATH-500, ScienceQA, MMLU-Pro, and GPQA.

With the increasing versatility of text-to-image diffusion models, the ability to selectively erase undesirable concepts (e.g., harmful content) has become indispensable. However, existing concept erasure approaches primarily focus on removing unsafe concepts without providing guidance toward corresponding safe alternatives, which often leads to failure in preserving the structural and semantic consistency between the original and erased generations. In this paper, we propose a novel framework, PAIRed Erasing (PAIR), which reframes concept erasure from simple removal to consistency-preserving semantic realignment using unsafe–safe pairs. We first generate safe counterparts from unsafe inputs while preserving structural and semantic fidelity, forming paired unsafe–safe multimodal data. Leveraging these pairs, we introduce two key components: (1) Paired Semantic Realignment, a guided objective that uses unsafe–safe pairs to explicitly map target concepts to semantically aligned safe anchors; and (2) Fisher-weighted Initialization for DoRA, which initializes parameter-efficient low-rank adaptation matrices using unsafe–safe pairs, encouraging the generation of safe alternatives while selectively suppressing unsafe concepts. Together, these components enable fine-grained erasure that removes only the targeted concepts while maintaining overall semantic consistency. Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art baselines, achieving effective concept erasure while preserving structural integrity, semantic coherence, and generation quality.

Standard Flow Matching (FM) models typically employ a pointwise loss function, effectively assuming that the velocity vector at each spatial position is pointwise decomposed supervision given the conditioning. We argue that this Pointwise Independence Assumption is suboptimal for image generation, where strong local correlations exist across space. Neglecting these dependencies leads to artifacts such as structural inconsistency. In this paper, we introduce Contextual Flow Matching (CFM), a generalized framework that enforces consistency not just on individual pixels, but on the contextual relationships within the velocity field. We formulate the training objective using a set of generalized Context Operators---akin to convolution kernels---that abstract relationship constraints. Specifically, we instantiate a Differential Operator to capture high-frequency motion dynamics and an Aggregate Operator to stabilize low-frequency structural evolution. Extensive experiments on ImageNet demonstrate that CFM significantly accelerates convergence and enhances generation fidelity by explicitly modeling spatial dependencies.

In-context reinforcement learning (ICRL) has emerged as an effective paradigm for test time adaptation to unseen tasks without parameter updates. However, existing ICRL methods can exhibit brittle and unstable adaptions, and the mechanisms underlying such adaptions remain poorly understood. We provide a policy-gradient view of ICRL and argue that relying on trajectory-level return feedback can lead to high-variance updates when test time rollout is limited. Motivated by the actor--critic principle, we propose Context Value Informed ICRL (CV-ICRL), which equips the in-context policy with an explicit context value: the expected discounted return conditioned on the current state and accumulated interaction context. CV-ICRL trains a value head and writes its prediction back into the context as a value token, enabling TD-style return targets for lower-variance policy updates. Experiments on the Dark Room, Minigrid, and Procgen testbeds show that CV-ICRL substantially improves the stability of test time adaptation and achieves higher returns across tasks and environments. The source code and data of this paper are available at https://anonymous.4open.science/r/CV-ICRL-D161.

Continual learning, the ability of AI systems to improve through sequential experience, has attracted substantial interest, but no high-quality benchmark exists to evaluate it. We introduce Continual Learning Bench (CL-Bench), the first difficult, expert-validated benchmark designed to measure whether LLM-based systems genuinely improve with experience. CL-Bench spans six diverse domains (software engineering, signal processing, disease outbreak forecasting, database querying, strategic game-playing, and demand forecasting), each validated by domain experts and designed so that tasks share a learnable latent structure (codebase layout, disease outbreak dynamics, opponent strategies) that a stateful system can discover online but a stateless one cannot. We evaluate frontier models across several agent architectures, from naive in-context learning (ICL) to dedicated memory systems, introducing a gain metric to isolate learning from prior capabilities. We find that these systems leave headroom for improved continual learning: agents frequently overfit to immediate observations or fail to reuse knowledge across instances, and dedicated memory systems do not fix this---in fact, naive ICL outperforms systems dedicated to memory management. CL-Bench is the first benchmark to evaluate continual learning across diverse real-world domains with expert-validated tasks and isolate online learning from underlying model capability, showing a need for better continual learning systems.

Cross-view UAV geo-localization involves three design choices: how to represent the UAV observation, how to represent the geo-referenced scene, and how to infer pose from their interaction. The dominant paradigm organizes the world as a discretized 2DoF candidate library, coupling scene representation to inference and bounding pose accuracy by sampling resolution. We decouple scene representation from pose inference by encoding the geo-referenced world as a continuously queryable semantic neural field, enabling localization to be formulated as explicit optimization over a continuous 4DoF pose space — parameterized by 2D ground position, in-plane heading, and viewing scale. Pose hypotheses are evaluated globally in a shared semantic feature space, bypassing the sampling bottleneck of retrieval. We learn 4DoF pose-sensitive observation representations under full 4DoF supervision provided by ContinuScenes, a multi-city benchmark with dense near-nadir UAV imagery and complete pose annotations. Inference proceeds hierarchically, progressing from broad pose-space coverage to adaptive concentration on high-compatibility regions. Preliminary experiments demonstrate that our method substantially outperforms prevailing localization paradigms, including discrete retrieval and absolute pose regression, across increasingly strict 4DoF pose criteria.

Real-world image degradation is often unknown, spatially non-uniform, and compositional, requiring all-in-one restoration models to adapt a single set of weights to diverse local corruption patterns without test-time degradation labels. Existing methods typically modulate a shared backbone with global prompts or degradation descriptors, or route features through predefined expert pools. However, compact global conditioning can bottleneck localized degradation evidence, while static expert routing may produce homogeneous updates or rely on unstable sparse assignments. We propose \textbf{Continuous Expert Assembly} (CEA), a token-wise dynamic parameterization framework for all-in-one image restoration. CEA employs a lightweight \textbf{Cross-Attention Hyper-Adapter} to probe intermediate spatial features and synthesize instance-conditioned low-rank routing bases and residual directions. Each spatial token then assembles its own residual update via dense signed dot-product affinities over the generated rank-wise components, avoiding external prompts, static expert banks, and discrete Top-$K$ selection. The resulting assembly rule also admits a linear-attention perspective, making its dense token-wise routing behavior transparent. Experiments on AIO-3, AIO-5, and CDD-11 show that CEA improves average restoration quality over strong prompt-, descriptor-, and expert-based baselines, with the clearest gains on spatially varying and compositional degradations, while maintaining favorable parameter, FLOP, and runtime efficiency.

Model distillation provides an alternative when computing resources are limited. How to transfer the reasoning ability from teacher model to student model influences the performance of model distillation. We develop a mathematical framework for reasoning distillation grounded in dynamical systems theory. By modeling the teacher model's residual stream as a discrete dynamical system, we define a precise notion of *orbital stability*: perturbations transversal to the teacher's reasoning trajectory should contract, while the tangential reasoning signal propagates without active suppression. We derive the **Contractive Restoring Flow (CRF)** loss from first principles and prove that, at the loss minimum, it achieves strict transversal contraction at a constant rate $1-\alpha$, leaves the tangential dynamics unconstrained (tangential agnosticism), and produces a global basin of attraction with bounded steady-state error within the linearisation regime. Empirically, reasoning-trajectory distillation with the CRF loss outperforms other distillation methods, and shows greater robustness on long-chain reasoning tasks, supporting the predicted effect of contracting transversal error. The code is avaiable at: https://anonymous.4open.science/r/CRF-5945/.

Visual reinforcement learning (RL) is promising for real-world problems such as autonomous driving, robotic locomotion, and manipulation, but remains highly vulnerable to visual distribution shifts between training and deployment. Existing methods improve robustness through Q-consistency regularization, masking, and auxiliary objectives, yet most of them follow a coupled representation learning paradigm in which control-oriented and robustness-oriented representation learning are jointly optimized within the same training framework. We argue that this coupled structure can introduce harmful interference between control learning and robustness learning, degrading original-environment control performance and limiting generalization under visual shifts. To address this issue, we propose Separate Then Align Representations (STAR), which reformulates visual RL generalization as a problem of decoupled representation learning. Specifically, we first learn a control-relevant reference representation through online RL under weak augmentation, and then perform an offline post-representation alignment stage that maps strongly augmented inputs back to this reference representation. Experiments on RL-ViGen benchmarks spanning DMControl and Robosuite show that STAR achieves strong generalization under visual distribution shifts while better preserving performance in the original environment than prior coupled approaches. Our source codes are available in the supplementary material.

Point cloud upsampling is essential for 3D applications but remains challenging due to sparse, under-constrained observations. Multi-view images can resolve geometric ambiguity, but existing multi-modal methods do not fully exploit visual cues. We propose ControlFlow3D, a framework that distills knowledge from pretrained foundation models into a lightweight generation network during training, enabling strong performance without multi-view images available at test time. Specifically, our method performs conditional flow matching in a structured latent space defined by a frozen 3D encoder. Multi-view image tokens and 3D latent tokens are jointly clustered into shared prototypes via unsupervised soft KMeans; the resulting prototypes guide the flow trajectory through velocity control and are aligned by a consistency loss that progressively transfers image knowledge into the base network. A Mamba-based geometry decoder then maps latent features to dense coordinates with linear complexity. We also construct ShapeNetPU, a 35K-object multi-modal benchmark that is 31times larger than prior datasets. Extensive experiments across multiple benchmarks demonstrate the effectiveness of our method and its strong generalization capability.

Next-token prediction leaves hidden-state trajectories largely unconstrained, motivating recent geometric regularization methods based on angular penalties. However, these approaches control only the direction of motion and provide no explicit bound on trajectory deviation. We introduce the Controlled Joint Embedding Predictive Architecture (ControlJEPA), a trajectory regularization framework grounded in discrete-time Input-to-State Stability theory. Our novel Lyapunov Tube Loss enforces a contraction condition on a normalized measure of transversal deviation, yielding a provable guarantee that confines hidden states within a tube of explicit width determined by two interpretable hyperparameters. ControlJEPA integrates seamlessly with standard training and requires no architectural changes. Across six benchmarks and seven model families, it consistently outperforms standard fine-tuning and prior trajectory regularizers, improving accuracy and data efficiency while preserving output diversity. Notably, the induced geometric structure persists at inference time, indicating that the method reshapes the representation space rather than memorizing training trajectories.

We present T2Mo, the first feed-forward framework for controllable dynamic 3D shape generation of general objects conditioned on 3D trajectories and text prompts. Existing text-driven methods struggle to specify precise motion due to the ambiguity of natural language. To address this, we introduce 3D trajectories as a direct spatial control signal and propose a shape-grounded trajectory embedding that maps arbitrary user-provided trajectories into geometry-aware condition tokens aligned with the input shape. Injecting these tokens into the generative backbone enables highly controllable motion generation while handling varying trajectory numbers and distributions. We conduct extensive comparisons against text-based baselines and trajectory-guided video generation workarounds. Quantitative and qualitative evaluations, along with user studies, show that our method produces motions that more faithfully follow the given prompts with higher expressiveness while preserving motion quality. Our code and weights will be publicly released.

Using offline datasets to evaluate conversational agents often fails to cover rare scenarios or to support testing new policies. This has motivated the use of \emph{controllable user simulators} for targeted, counterfactual evaluation, typically implemented by prompting or fine-tuning large language models. In this work, we formalize controllable simulation as a causal inference problem. By bridging natural language evaluation with off-policy evaluation methodology, we show that the standard practice of training simulators via supervised fine-tuning on post-hoc trajectory labels yields a structurally biased model. Specifically, these labels are inextricably coupled to the data-generating behavior policy, injecting a \emph{look-ahead bias} that breaks causal consistency. Furthermore, we prove that under policy shift this failure causes the variance of evaluation metrics to explode geometrically, a phenomenon we term \emph{controllability collapse}. To restore causal consistency, we establish theoretical conditions for accurate simulation and propose practical training mitigations: \textit{a priori} controls, step-wise dynamic controls, and direct policy-conditioned learning. Empirical evaluation confirms that while standard global controls distort conversational distributions and collapse behavioral diversity, our causally grounded simulators eliminate look-ahead bias, preserve natural variance, and exhibit robust zero-shot generalization to unseen agent behaviors.

Autoregressive (AR) models have reformulated Scalable Vector Graphics (SVG) generation as next-token prediction, demonstrating remarkable potential in text-to-SVG tasks. However, controllable SVG generation guided by spatial signals (e.g., sketches or scribbles) remains largely unexplored within AR models. While a natural approach, inspired by controllable image generation, is to adapt methods like Condition Prefilling or Conditional Decoding, they fall short in SVG generation. To address these challenges, we introduce ControlSVG, a versatile and effective framework tailored for integrating precise spatial controls into autoregressive SVG generation. Firstly, we propose a Visual-Feedback Conditioning mechanism that constructs a “See-and-Draw" loop. This design mitigates cross-modal drift and ensures long-range consistency. Secondly, we explore spatially aware learning objectives that explicitly bridge discretized SVG tokens with their absolute 2D spatial coordinates, enhancing robustness against coordinate deviations during sequential generation. Extensive experiments on Canny-Edge-to-SVG generation demonstrate that ControlSVG outperforms existing baselines, achieving superior spatial alignment and high-quality SVG. We further provide preliminary validations on real human-drawn scribbles and extend the framework to scribble-to-SVG and image-to-SVG settings, suggesting its potential applicability beyond Canny-based control.

Nanodevices that absorb energy from their thermal environment can act as quantum refrigerators. Finding suitable configurations, however, is a needle-in-a-haystack problem: the many-body state space scales exponentially with the size of the device, where each candidate configuration requires an expensive quantum-chemical computation. Under random sampling of material parameters, only 2-9\% of configurations exhibit the desired property. We introduce a search framework that couples a Graph Variational Autoencoder (GVAE) with an uncertainty-driven Active Learning (AL) loop to efficiently identify energy-absorbing configurations in electronic junctions with a 2D "working material" to be optimized. The GVAE maps physical parameters - site energies, coupling matrices, temperature - into a smooth latent space that distinguishes cooling from dissipative parameter regimes, achieving a classification ROC-AUC of 0.89 on held-out topologies and 0.80 on unseen extrapolation topologies. An AL pipeline based on Monte Carlo dropout then targets the decision boundary, iteratively selecting the candidates where the model is least certain for simulation. Over 32 AL cycles, the uncertainty-driven strategy discovers positive-flux (energy absorbing) systems at rates consistently exceeding the random baseline across all tested topologies, providing a data-efficient pathway for navigating a combinatorically large design space. The learned latent space organizes configurations along physically interpretable axes, that correlate with underlying microscopic model parameters. Our framework, thus, paves a new way towards effective cooling of computational devices, to reduce the environmental footprint of high performance computing.

Mode connectivity studies reveal that high-performing neural networks are often connected by low-loss paths in parameter space. We investigate the analogue of this phenomenon in cooperative multi-agent reinforcement learning (MARL), where a solution is not a single model but a team of decentralized policies. We introduce coordination connectivity, a diagnostic evaluating whether linear interpolation between the actor parameters of two trained policy teams preserves high team return. The resulting coordination barrier measures whether two successful teams are separated by a performance cliff along the joint-policy path. We evaluate this diagnostic on the StarCraft Multi-Agent Challenge (SMAC) across five controlled training histories, isolating the effects of shared initialization, continued training, parameter decoupling, and HAPPO-style separate-actor specialization. Across three SMAC maps and four seeds, policy teams derived from a shared MAPPO checkpoint exhibit near-zero interpolation barriers, even after non-shared specialization. In contrast, interpolation paths involving independently trained HAPPO teams exhibit substantially larger barriers. Mixed-team assembly, single-agent hot-swap, clone-all, role-swap, and weight-space probes further indicate that shared-initialized agents specialize into distinct roles while remaining within a low-barrier coordination component. These results suggest that shared initialization implicitly structures the joint-policy space of cooperative MARL, providing a mode-connectivity-inspired lens for analyzing coordination, specialization, and same-source cross-version team compatibility.

LLM coding agents operate by constructing trajectories that accumulate reasoning, tool calls, and results to enable multi-step decision-making. However, the conventional append-only trajectory architecture found in practice tightly couples file-read actions with their observations, capturing snapshots that become permanently fixed in the chronological history. As files change through agent edits or concurrent human modifications, these snapshots become stale, causing reasoning errors and causing agents to redundantly re-read files, with each re-read appending yet another copy to the trajectory. To mitigate this, we propose CORVUS, a novel trajectory architecture that decouples file-read actions from their observations by maintaining a synchronized registry of relevant files and injecting only their current contents at each reasoning cycle. This structural change produces significantly lighter-weight trajectories that remain synchronized with the actual codebase state by construction, eliminating redundant file copies and stale snapshots that bloat conventional trajectories. We evaluated CORVUS on SWE-POLYBENCH_VERIFIED and SWE-BENCH PRO across four LLMs, achieving 9–50% reduction in average input tokens per task, 15–32% shorter final prompts, and up to 37% fewer reasoning cycles while maintaining comparable pass rates.

Monitoring the chain-of-thought (CoT) of reasoning models is a promising approach for detecting covert misbehavior (i.e., hidden objectives) in code generation tasks. While large models (GPT-5, Gemini-3-Flash) can serve as effective CoT monitors, they are expensive to deploy due to the lengthy reasoning traces and high API cost, emphasizing the need for smaller, cheaper alternatives. Nevertheless, we find that current small models (4B--8B) struggle to detect hidden objectives despite access to the CoT, frequently misattributing them as part of the user query. To address this, we propose a post-training pipeline combining supervised fine-tuning (SFT) and reinforcement learning (RL), where SFT narrows the gap for in-domain tasks by distilling detection behavior from stronger monitors, and RL on hard and subtly crafted hidden objectives helps the model generalize to out-of-domain monitoring tasks. To validate this generalization, we evaluate under a realistic threat model motivated by practical supply-chain attacks, where the adversary is a third-party LLM router injecting hidden objectives into code-generation requests through either prompt manipulation or code manipulation attacks. To push beyond objectives that large monitors already saturate, we also introduce four new challenging tasks even for strong monitors. Finally, we introduce CoT-Guard, a 4B-parameter monitor that demonstrates superior generalization performance under both prompt and code manipulation attacks, achieving a $\text{Gmean}^2$ (i.e., TNR$\times$TPR) of 75\% and outperforming GPT-5.4 (56\%), GPT-5-mini (41\%), and Qwen3-32B (54\%), while closing the gap to Gemini-3-Flash (83\%). These results demonstrate that \modelname provides a practical and cost-effective user-side defense, substantially improving hidden-objective detection while avoiding the deployment cost of large monitors.

Estimating counterfactual outcomes when multiple treatments are applied simultaneously, the composite treatment setting, is fundamentally harder than the single-cause case: the treatment space grows combinatorially, most combinations are unobserved, and confounding is entangled across dimensions. We propose PACE (Progressive Alignment for Composite Effects), which exploits a known causal DAG over treatment variables to progressively transform the observational data distribution into the target interventional distribution. PACE performs K sequential augmentation steps in reverse topological order, each isolating a single treatment component and conditioning only on its parents in the DAG. We derive the closed-form optimal augmentation density minimizing KL divergence at each step, establish finite-sample bounds prescribing the augmentation budget via DAG-aware selection bias ratios, and show that the DAG provides a deterministic processing order that eliminates the need for random permutation averaging. On semi-synthetic benchmarks with real covariates (IHDP, Twins, Hillstrom) and controlled synthetic environments, PACE substantially outperforms seven baselines on both continuous and binary outcomes, with the largest gains on rare treatment combinations where confounding is most severe. Ablations confirm consistent improvements over the closest prior method across varying DAG topologies, confounding strengths, and under DAG misspecification.

Counterfactual explanations (CFs) provide human-interpretable insights into model's predictions by identifying minimal changes to input features that would alter the model's output. However, existing methods struggle to generate multiple high-quality explanations that (1) affect only a small portion of the features, (2) can be applied to tabular data with heterogeneous features, and (3) are consistent with the user-defined constraints. We proposeCounterFlowNet, a generative approach that formulates CF generation as sequential feature modification using conditional Generative Flow Networks (GFlowNet). CounterFlowNet is trained to sample CFs proportionally to a user-specified reward function that can encode key CF desiderata: validity, sparsity, proximity and plausibility, encouraging high-quality explanations. The sequential formulation yields highly sparse edits, while a unified action space seamlessly supports continuous and categorical features. Moreover, actionability constraints, such as immutability and monotonicity of features, can be enforced at inference time via action masking, without retraining. Experiments on eight datasets under two evaluation protocols demonstrate that CounterFlowNet achieves superior trade-offs between validity, sparsity, plausibility, and diversity with full satisfaction of the given constraints.

Building hydrogen station networks, electric vehicle (EV) charging grids, or vaccine distribution systems requires coordinating two separately operated graphs - demand-side facilities and the supply-side networks they depend on - where each operator's payoff depends on the other's choices. We formalize this as the Co-Evolving Graph Game (CEGG), a setting that sits between facility location, interdependent-network analysis, and graph-based multi-agent reinforcement learning (MARL) but is not jointly addressed by any of them. We propose Cross-Graph Attention with Policy Mirror Descent (CGA-PMD), pairing cross-graph attention - which lets each agent read only the partner-graph nodes structurally coupled to its own - with a policy mirror descent update for stable training. Across three reward domains and a range of practical network scales, we find that the structure of the coupling selects which method wins: at small scale most methods are competitive, but as scale grows or coupling becomes irregular, only CGA-PMD remains productive while own-only, unmasked-attention, and Proximal Policy Optimization (PPO)-based variants all fail in distinct ways. Our theoretical analysis explains why: PMD's Kullback-Leibler (KL) regularization admits an iteration-uniform cumulative drift bound while PPO's clip update does not, predicting CGA-PMD's stability in the hard regime where PPO-based methods collapse.

Reproducing machine learning research is challenging because many papers do not provide executable code, and key implementation details are often scattered, implicit, or missing. Existing paper-to-code systems improve over direct prompting, but the generated repositories can still miss paper-critical logic or fail at execution time. We propose Crafter, an automated paper-to-code pipeline that treats reproduction as a problem of context-calibrated implementation recovery rather than single-pass code generation. Crafter builds an evidence-grounded specification, resolves missing implementation details before planning, and repairs the generated repository through execution-aware debugging. We evaluate Crafter with PaperBench Code-Dev for implementation faithfulness and a self-designed execution-oriented benchmark for runtime readiness. Across 23 PaperBench papers using the Claude Sonnet 4.6 backend, Crafter achieves an average Code-Dev score of 0.84, improving over Paper2Code by 19.7% and over DeepCode by 22.1%. In execution evaluation, Crafter reaches the contribution-level milestone on all three repositories, compared with two for both baselines, while reducing average repair cycles from 11.0--11.3 to 5.0.

Traditional forecasting evaluations relying on static historical data are fundamentally incompatible with modern AI advances. Even for standard time series foundation models, temporal overlap between training and test periods can lead to implicit leakage through correlated signals. This problem is intensified by LLMs, which can bridge semantic gaps between forecasted and related data while memorizing extensive world knowledge. At the extreme, agentic models with tool access can retrieve the ground truth directly. Such information leakage inflates performance relative to real forecasting scenarios and is impossible to rule out in historical evaluations. We introduce CREF, a real-time benchmark that evaluates forecasting models on data that does not yet exist, eliminating all forms of leakage by design. CREF balances three competing objectives for real-time benchmarks: diversity, robustness, and throughput. It draws data from 32 public APIs and features 50 forecasting tasks with diverse data frequencies, multivariate targets, and covariates, making it the first real-time benchmark reaching the breadth of leading static benchmarks. Retrospective evaluation validates that CREF produces rankings consistent with established static benchmarks, while real-time evaluation reveals qualitative and quantitative performance gaps for agentic and LLM-based models, confirming leakage on historical data. CREF runs continuously and is open for new models to join, offering a public leaderboard for contamination-free model comparison.

Croissant has emerged as the metadata standard for machine learning datasets, providing a structured, JSON-LD-based format that makes dataset discovery, automated ingestion, and reproducible analysis machine-checkable across ML platforms. Adoption has accelerated, and NeurIPS now requires Croissant metadata in every submission to its dataset tracks. Yet in practice Croissant generation usually starts with uploading data to a public platform, a path infeasible for governed and large local repositories that hold much of the high-value data ML increasingly relies on. We release Croissant Baker, a local-first, open-source command-line tool that generates validated Croissant metadata directly from a dataset directory through a modular handler registry. We evaluate Croissant Baker on over 140 datasets, scaling to MIMIC-IV at 886 million rows and 374 Parquet files. On held-out comparisons against producer-authored or standards-derived ground truth, Croissant Baker reaches 97–100% agreement across multiple domains.

Multimodal geometry reasoning requires models to jointly perceive diagrams and perform symbolic derivations. Test-time verification methods such as majority voting assume voter errors are roughly independent, but in geometry this assumption breaks down: all candidates perceive the same diagram, so perceptual biases cascade into correlated errors that produce confident but wrong consensus. We observe that neuro-symbolic candidates contain compositional structure that can serve as a verification resource. We introduce SKETCH, a format decomposing each candidate into a neural pathway (natural-language claim) and a symbolic pathway (visual grounding → typed DSL → deterministic execution). These pathways are governed by different error processes—symbolic errors are correlated across candidates through shared perception, while neural errors are more dispersed. Our method, Compositional Consensus Verification (CCV), gates consensus on cross-channel agreement: because the two pathways fail for different reasons, accidental agreement is rare, making each concordant vote far more precise than a raw majority vote. Across four geometry benchmarks, CCV achieves 81.0% accuracy (+10.4 pp over majority voting) without external scorers or additional inference. Seven non-compositional baselines—including trained PRMs—cluster at 70–71%, while compositional cross-layer features jump to 79–81%, confirming that structural diversity is the core verification resource, and that a principled gating algorithm effectively unlocks its potential.

Visual Foundation Models (VFMs) have demonstrated strong capabilities in visual-geometric tasks such as image matching and 3D reconstruction. Recent work has revealed the potential of VFMs in label-free indoor RGB-D registration, where the visual modality is intrinsic to the task. A natural question follows: can this success extend to more challenging outdoor scenarios characterized by sparse, irregular LiDAR data with only partial visual coverage? We answer this with PROMOTER, a novel teacher–student framework for unsupervised LiDAR registration that bridges visual priors and sparse LiDAR geometry. To leverage pre-trained geometric priors for 3D registration, we must disentangle them from confounding signals introduced by modality and task gaps under partial visual coverage. To this end, we introduce a lightweight Adaptive Prior Integrator that extracts and integrates these priors into an enhanced teacher, together with a Perceptual-Geometric Labeler that produces high-quality pseudo-labels. For student training, we propose Matchability-Anchored Training to improve robustness under noisy supervision. Experiments on KITTI and nuScenes demonstrate state-of-the-art performance and improved scalability across registration models and VFMs, while incurring no additional parameters at inference. Code will be released.

Despite remarkable advances, today's AI systems remain narrow in scope, falling short of the flexible, adaptive, and multisensory intelligence that characterizes human capabilities. This gap has fueled longstanding debates about whether AI might one day achieve human-like generality or even consciousness, and whether theories of consciousness can inspire new architectures for AI. This paper presents an early blueprint for implementing a general AI system, CTM-AI, combining the Conscious Turing Machine (CTM), a formal machine model of consciousness, with today's foundation models. CTM-AI contains an enormous number of powerful processors ranging from specialized experts (e.g., vision-language models and APIs) to unspecialized general-purpose learners poised to develop their own expertise. Crucially, for whatever problem must be dealt with, information from many processors is selected, integrated, and exchanged appropriately to solve the task.CTM-AI achieves state-of-the-art accuracy on MUStARD (72.28) and UR-FUNNY (72.13), outperforming multimodal and multi-agent frameworks. On tool-using and agentic tasks, CTM-AI achieves 10+ points of improvement on StableToolBench and WebArena-Lite. Overall, CTM-AI offers a principled, testable blueprint for general AI inspired by a model of consciousness.

Flow Matching (FM) has emerged as the dominant paradigm for state-of-the-art image and video generation, yet its reliance on full attention over lengthy sequences incurs significant latency. While existing caching methods reduce redundant computations, they yield suboptimal performance, fundamentally constrained by inflexible spatial granularity and empirical, model-specific heuristics. To bridge this gap, we ground the caching mechanism in the geometric properties of the FM velocity field. We theoretically establish that the \textbf{curvature} inherently dictates the coupling between trajectory fluctuation and computational requirements, thereby providing a rigorous metric to quantify caching staleness loss and automatically derive adaptive granularity and optimal update intervals. Translating this theoretical insight into practical acceleration, we present CuBic, a train-free caching framework. It bypasses expensive curvature computations via a zero-overhead proxy signal that dynamically guides spatial partitioning and budget-constrained update scheduling. At the system level, a sparse forward engine translates these optimal policies into efficient runtime, ensuring that theoretical guarantees directly translate into actual FLOP savings and linear speedup. Being model-agnostic and virtually hyperparameter-free, CuBic consistently outperforms state-of-the-art caching baselines in quality at customizable acceleration ratios (averaging 2.5$\times$) across Wan2.1, Flux.1, and FireRed-Image-Edit.

Diffusion language models (DLMs) generate text through iterative denoising, enabling parallel decoding, bidirectional conditioning, and editable intermediate states. Under harmful prompts, however, intermediate denoising states can contain risk-inducing tokens that are seemingly benign but increase the probability of an unsafe final response by shaping how the remaining masked positions are completed. Since committed tokens in DLMs can still be re-masked and rewritten, safety control can be applied during denoising by removing a small set of risk-inducing tokens before they drive subsequent generation toward unsafe responses. Existing DLM defenses mainly rely on model-level alignment, trajectory-level detection, or block-level repair, often requiring training cost, extra inference passes, or coarse regeneration. We propose CURE, a counterfactual value-driven test-time alignment method that keeps the base DLM fixed and selectively re-masks risk-inducing tokens. CURE trains a time-conditioned safety value model to estimate whether a partially denoised state will lead to an unsafe final response. During inference, CURE constructs counterfactual readout views that remove or isolate targeted tokens, estimates their contribution to future unsafe response, and re-masks only high-risk tokens for later rewriting. Across three DLMs and five jailbreak benchmarks, CURE reduces macro-average ASR from 47.20\% to 1.36\%, preserves utility on MMLU and GSM8K, and adds only 0.84 extra TFLOPs, achieving a strong safety-utility-efficiency trade-off. Code is available at \url{https://anonymous.4open.science/r/CURE}.

Direct Preference Optimization is a widely used approach for safety alignment, aiming to reduce harmful behaviors in large language models. However, prior work shows that it can be brittle and exhibits poor out-of-distribution (OOD) generalization \citep{qi2025safety}. To this end, in this paper we investigate whether curriculum learning can improve the robustness of DPO-based safety alignment. We propose \textbf{Staged-Competence}, a curriculum-based framework that organizes preference data by difficulty, employs competence-based sampling, and progressively updates the reference model during training. Averaged across three model families, Staged-Competence reduces OOD harmful response rates by 16\% and jailbreak attack success rates by 20\%, while preserving general capabilities and maintaining near-zero over-refusal. We further show that Staged-Competence (1) matches baseline safety with only 75\% of the training data, demonstrating improved data efficiency and (2) yields better separation between safe and unsafe responses. Staged-Competence is agnostic to the underlying policy optimization loss and can extend to other DPO variants and alignment domains other than safety. Our code and data can be found at: \url{link/upon/acceptance}.

Treatment assignment problems arise wherever limited budget must be allocated to heterogeneous users, with applications ranging from personalized recommendations to online advertising and healthcare. In such settings, individuals exhibit heterogeneous responses to different treatments, making it essential to learn cost-aware personalized treatments. This paper introduces the Cost per Unit Value Equalization Tree (CUVET) algorithm, a novel treatment assignment approach that partitions the user space. Under a diminishing-returns (power-law) assumption, CUVET solves the within-cohort allocation problem by equalizing the marginal cost per unit value across each user group, yielding a closed-form cost-aware treatment assignment suited for large-scale industrial deployment. We also release CUVET-policy, an 86.7-million-impression public benchmark derived from real-world industrial A/B tests, providing an open-source evaluation framework for decision-focused learning. Across a multi-method benchmark suite, the Bayesian variant of CUVET achieves the largest cost-feasible value uplift on the public MT-LIFT dataset +11.1% under LP, vs. at most +1.3% for all baselines, while on CUVET-policy, where treatment effects are sub-percent, CUVET variants are the most efficient and strongest cost-feasible methods.

Unlike chunk-based Retrieval-Augmented Generation (RAG) methods which retrieve information only from text passages, graph-based RAG methods improve their performance through entity-relation graphs. However, existing graph-based RAG methods typically represent graph facts under the Open World Assumption (OWA), where an absent edge (relation) means unknown rather than false. This principle is problematic for many closed corpora including technical manuals and regulations, where rules are generally stated through scope-bounded statements (such as all X, except Y or only Y) named as local completeness statements(LCSs) in this paper. LCS apply rules to every unmentioned in-scope entity which cannot be explicitly and completely represented by the OWA graph. To address this problem, we propose CWAGraph, which is a graph-based RAG framework augmented with a Closed World Assumption (CWA) layer representing local scope and boundaries for LCSs. The CWA layer identifies which entities are governed by each statement and makes the corresponding closed-world evidence retrievable from those entities, thus enhancing the performance of question answering (QA) involving LCS (called as LCS resolution in this paper). To better evaluate RAG systems' performance on LCS resolution, we further introduce a new QA benchmark SetQA consisting of 493 QA instances involving LCSs across two source domains and three difficulty levels. Our experiments on SetQA reveal that CWAGraph consistently outperforms chunk-based and graph-based RAG baselines.

We present CyCLeGen, an autoregressive framework that integrates layout understanding and layout-to-image generation through cycle consistency: predicted layouts must produce faithful images, and generated images must yield consistent layouts. We enforce this constraint via CycleGRPO, a bidirectional reinforcement learning strategy with complementary geometric and perceptual rewards, enabling self-introspective learning from only 8k RL samples. This creates a natural loop, generation helps understanding by rewarding only those layouts that lead to high-quality images, and understanding helps generation by rewarding only those images whose spatial structure can be faithfully recovered. Extensive experiments show that CyCLeGen achieves significant gains across diverse image understanding and generation benchmarks, with emergent gains on image captioning.

Offline multi-task reinforcement learning (Offline MTRL) heavily depends on the quality and distribution of pre-collected data. However, existing methods mainly focus on algorithmic optimization, with less emphasis on data-level improvements to enhance learning ability and generalization performance. This paper, from a data perspective, reveals three key bottlenecks that limit Offline MTRL performance:(i) ineffective utilization of prompts length under diverse task complexities, and (ii) semantic irrelevance of randomly sampled prompt segments, (iii) misleading supervision induced by fragmented and discontinuous trajectories. To address these challenges, we propose DaCe-DT, a robust offline MTRL framework designed to be insensitive to heterogeneous task complexities and data quality, featuring length-gated prompt masking (LGPM), retrieval-augmented prompt construction (RAPC), and value-adaptive return calibration (VARC). Together, these mechanisms enable DaCe-DT to deliver data-centric prompt adaptation and trajectory refinement, resulting in robust multi-task generalization and stable policy learning amid heterogeneous offline data and tasks. Experimental results on Meta-World show that DaCe-DT consistently outperforms state-of-the-art methods, achieving an average improvement of 11.73\% on optimal datasets and an improvement of 13.34\% on suboptimal datasets, demonstrating its effectiveness in learning stably from imperfect data and improving overall multi-task performance.

Pathological whole slide images (WSIs) from different hospitals often exhibit severe domain shifts due to variations in scanning devices, staining protocols, and tissue preparation procedures. This leads to significant performance degradation when a trained model is applied to data from different sources. Although unsupervised domain adaptation (UDA) has achieved strong performance in transferring a model trained on a source domain to an unlabeled target domain in natural images, it is much less studied in the computational pathology field. This is due to the unique properties of WSIs, including ultra-high resolution and strong inter-slide and intra-slide heterogeneity. As a result, existing UDA methods are often unstable and may even cause negative transfer in WSI classification. To address this issue, we propose Dual-level Adversarial learning with domain-aware REgularization (DARE) for WSI Classification. Specifically, we design an adaptive pseudo-labeling framework with a teacher-student architecture to provide stable supervision for the target domain and perform adversarial alignment on both patch-level and slide-level embeddings. This enables joint optimization of patch-level and slide-level semantic embeddings for dual-level alignment across domains. We also introduce domain-aware attention masking to regularize attention and improve cross-domain generalization. We evaluate our method on four public WSI datasets through extensive experiments. The results show that our method consistently outperforms state-of-the-art approaches across various domain shifts.

As the ecosystem of Large Language Models (LLMs) rapidly expands, LLM routing has become essential for dynamically balancing performance and cost. However, existing methods fail to simultaneously satisfy the three core criteria of an ideal router, i.e., high routing precision, minimal intrinsic operational overhead, and zero-shot onboarding of new models. These methods rely heavily on surface-level semantics and discrete ID mappings, conflating mere semantic similarity with underlying capability requirements and thereby precluding zero-shot, training-free generalization. To address these limitations, we propose DART, a novel zero-shot routing framework grounded in dual-side alignment to optimize LLM performance-cost tradeoffs. Instead of relying on direct semantic matching, we decouple the routing process into a demand-supply alignment problem within a shared low-rank latent capability space. Specifically, on the supply side, DART constructs latent capability embeddings for candidate models, using an initialization mechanism to anchor newly introduced models in this latent space via metadata and family-graph prior, without retraining. On the demand side, a dedicated encoder explicitly distills the query's directional capability requirements and inherent difficulty into a latent demand embedding. This decoupled representation allows routing decisions to be computed via a lightweight dot-product operation between the demand and capability embeddings, modulated by query-adaptive cost sensitivity, minimizing overhead. Extensive experiments across comprehensive benchmarks demonstrate that DART achieves state-of-the-art utility under diverse cost constraints, satisfying the three core criteria with exceptional robustness in cold-start scenarios.

Symbolic regression (SR) aims to discover interpretable mathematical expressions from data and plays a key role in scientific discovery. However, existing methods face a common bottleneck: the enormous search space and severe combinatorial explosion make it difficult to achieve a favorable trade-off between accuracy and algorithmic efficiency. We argue that, in addition to designing increasingly complex solver-specific SR strategies, an equally promising direction is to equip them with a shareable auxiliary strategy that can provide reliably search guidance. To this end, We propose DASS, a solver-agnostic Dynamic Auxiliary Search Strategy that improves SR by constructing and refining a high-potential auxiliary search subspace. DASS represents the auxiliary search space as a functional subspace spanned by a set of high-potential basis function terms. It estimates term utility through multi-environment evaluation, filters unstable guidance, and updates terms' potential score with a Gibbs posterior, thereby calibrating the subspace to provide more reliable guidance for downstream solvers. The High-quality solutions produced by the solver are used to update the set of basis terms. This forms a virtuous closed-loop optimization process: the subspace guides the downstream solver, and the solver feedback reshapes the subspace for subsequent exploration. DASS can be seamlessly integrated with various types of SR solvers. Extensive experiments on LLM-SRBench show that DASS significantly improves the accuracy of original solvers with a manageable increase in runtime.

In this paper, we propose Data-Adaptive Mahalanobis Cross-Head Attention (DAMCHA), a simple and effective attention mechanism that generalizes Multi-Head Attention (MHA) through the lens of metric learning. Standard MHA implicitly learns a static, head-wise Euclidean similarity corresponding to the diagonal blocks of a unified cross-head metric matrix, discarding all off-diagonal interactions across heads. DAMCHA addresses this limitation via two key innovations. First, it learns the full cross-head metric matrix, inducing a Mahalanobis-like similarity that explicitly captures inter-head interactions beyond the restricted block-diagonal structure of MHA. Second, it parameterizes the metric matrix as an input-dependent function, enabling the attention geometry to dynamically adapt to the intrinsic manifold structure of the data rather than remaining fixed after training. We further introduce principled regularization strategies, most notably stack-wise parameter sharing, to ensure computational efficiency and stable optimization. DAMCHA serves as a drop-in replacement for MHA and its variants in various Transformer architectures, delivering superior expressive power and learning performance with comparable or reduced computational overhead. Extensive experiments demonstrate that DAMCHA-based Transformers consistently outperform competing baselines across a range of benchmark tasks. Code is available at https://anonymous.4open.science/r/DAMCHA-3B37.

Building performant Vision-Language Models (VLMs) requires carefully curating large-scale training datasets, yet the community lacks systematic benchmarks for evaluating such curation strategies. We introduce DataComp for VLMs (DCVLM), a benchmark for controlled data-centric experiments to improve VLM training. As part of DCVLM, we standardize 160 datasets spanning four data types—image-caption pairs, multimodal interleaved documents, text-only, and instruction-tuning data—into a corpus of 6T multimodal tokens. DCVLM allows participants to test curation strategies (filtering, mixing, formatting, sampling) across 1B–8B models and 6.25B–200B token budgets. Models are then evaluated on a carefully selected suite of up to 52 downstream benchmarks across 9 domains. We conduct extensive experiments on DCVLM and find that data mixing, not filtering, is key to a high-quality training dataset: instruction-heavy mixtures scale better than caption-heavy ones, with gains widening at larger scales. The resulting dataset, DCVLM-BASELINE, enables training an 8B VLM to 63.6% accuracy on our 33-task core suite with 200B training tokens. Compared to FINEVISION, the state-of-the-art open VLM training dataset, this represents an improvement of +5.4pp. DCVLM and all accompanying artifacts will be made publicly available.

Constraint satisfaction problems (CSPs) serve as critical benchmarks for neural algorithmic reasoning, requiring accurate multi-step constraint propagation to determine variable values that satisfy all specified constraints. Existing neural solvers often rely on extensive training datasets, struggle with distribution shifts, or allow fixed clues to drift in continuous spaces. We introduce CSP-RSB, a data-efficient recurrent Transformer that directly incorporates constraint topology into the attention mechanism through a learnable relational bias. Additionally, we propose a hard axiom clamping strategy that fixes clues after each recurrence constraint projection and uses the QK-Norm to stabilize a two-layer block. In 9 \times 9 Sudoku, CSP-RSB achieves 99.94\% board accuracy on the RRN-test, 99.13\% on unfiltered Kaggle, and 99.78\% on filtered Kaggle, surpassing the previous best, Trial &amp; Error 2025, which reported accuracies of 99.40\%, 98.90\%, and 99.50\%, respectively. Furthermore, it successfully solves Erd\H{o}s--R\'enyi graph coloring with 100\% accuracy in 9 epochs and maintains 99.5-100\% accuracy on 10 \times 10 binary puzzles. By incorporating explicit topological structure, CSP-RSB reduces the need for depth and data requirements for effective algorithmic reasoning.

Cold-start exemplar-free class-incremental learning requires learning a growing set of classes without replay, external pretraining, or a large initial task. Existing cold-start methods typically either train the backbone throughout the stream and compensate for semantic drift, or freeze a backbone after the first task, producing features biased toward the initial classes. These choices also create a computational tension: drift-compensation methods require repeated backbone training and increasingly expensive updates as the task horizon grows, while frozen-backbone methods are cheap but weak under cold start. We study a third option: a feature extractor that is never fit to image data at all. We propose CIRCLE, a class-incremental classifier built from fixed bidirectional two-dimensional reservoir features, adapted from BiRC2D for image classification, and streaming linear discriminant analysis heads. CIRCLE groups multiple random reservoir instantiations into feature ensembles and averages the softmax outputs of independent SLDA heads, yielding a tunable bias-variance tradeoff between richer random features and prediction-level ensembling. Because the feature extractor is fixed and the head admits streaming closed-form updates, CIRCLE performs sample-wise training without replay, task-boundary information, or backbone backpropagation. On CIFAR-100, TinyImageNet, ImageNet-Subset, and ImageNet-1k, CIRCLE is competitive at 10-20 task splits and substantially outperforms strong CS-EFCIL baselines at 50, 100, and 500 task splits, while training much faster than trained-backbone drift-compensation methods. Ablations show that the BiRC2D-style extractor, SLDA head, and balanced feature/prediction ensembling each contribute to the final performance.

The task of \textit{dataset distillation} aims to find a small set of synthetic images such that training a model on them reproduces the performance of the same model trained on a much larger dataset of real samples. Existing distillation methods focus on synthesizing datasets that enable training \textit{randomly initialized} models. In contrast, state-of-the-art vision approaches are increasingly building on large, pre-trained self-supervised models rather than training from scratch. In this paper, we investigate the problem of distilling datasets that enable us to optimally train \textit{linear probes} on top of such large, pre-trained vision models. We introduce a method of dataset distillation for this task called \textit{Linear Gradient Matching} that optimizes the synthetic images such that, when passed through a pre-trained feature extractor, they induce gradients in the linear classifier similar to those produced by the real data. Our method yields synthetic data that outperform all real-image baselines and, remarkably, generalize across pre-trained vision models, enabling us, for instance, to train a linear CLIP probe that performs competitively using a dataset distilled via a DINO backbone. Further, we show that our distilled datasets are exceptionally effective for fine-grained classification and provide a valuable tool for model interpretability, predicting, among other things, how similar two models' embedding spaces are under the platonic representation hypothesis or whether a model is sensitive to spurious correlations in adversarial datasets.

High-compression tokenizers are essential for scaling latent image generative models. However, aggressive compression creates a fundamental tradeoff between reconstruction fidelity and generation efficiency: high compression image encoder always increase the learning diffucity of diffusion training, resulting slow model convergence. Recent representation autoencoders speed-up the diffusion training by improving the latent feature's expressive capability by replacing VAE encoders with pretrained semantic encoders, yet they are typically limited to moderate compression and lose pixel-level details necessary for faithful reconstruction. To achieve both high compression and fast diffusion training, we propose DC-SAE, a Decoupled Compact Semantic Autoencoder designed for high-compression image generation with accelerated diffusion model convergence. DC-SAE consists of two key components: (1) a macro-level architecture design that leverages semantic encoders to enable higher compression ratios, and (2) a pixel-level encoder that preserves low-level details, ensuring high-fidelity image reconstruction. We empirically demonstrate that DC-SAE performs strongly on image generation tasks, achieving both compact latent representations and efficient training dynamics. % Specifically, on ImageNet dataset with $512 \times 512$ resolution, DC-SAE achieves $32\times$ spatial compression, with \textbf{29.79} PSNR and \textbf{3.05} gFID, substantially outperforming the previous state-of-the-art high-compression tokenizer baselines DC-AE by 13.5\% and 59.7\% on PSNR and gFID respectively, maintaining comparable throughput and faster diffusion model training convergence.

Foundational visual features such as DINO have played a critical role across modern computer vision, and have recently become key components in multi-view feed-forward geometry estimators. In this work, we demonstrate that by re-distilling these multi-view models-their internal knowledge of 3D geometry-into a single-view estimator, we can obtain enhanced 3D consistent foundational features. Our key idea is to construct a multi-view teacher by fusing pretrained 2D foundation features with multi-view geometric features, and refining the fused representation with a discriminative ranking objective. Through our discriminative distillation framework, we enforce the learned features to be both 3D consistent and locally distinctive, while keeping them aligned with the feature space of the original foundation model to preserve the semantic structure of the pretrained representation. Consistency and local discriminability are critical for 3D computer vision problems such as forming semantic and geometric correspondences across images. To demonstrate the effectiveness of our method, we perform comprehensive experiments spanning multiple angles: direct feature analysis, dense prediction transfer, and explicit 3D lifting and rendering. Across these evaluations, our method consistently produces stronger 3D-aware foundation features that improve multi-view consistency and local discriminability while preserving the semantic transferability of the original representation.

Investigating the causal structures of transformers often entails a trade-off between the granularity of the explanation and computational overhead. Existing approaches either bypass the full internal chain, potentially missing critical decision-relevant links, or struggle with the combinatorial explosion of path enumeration. To address these challenges, we propose an efficient automated path-tracing method that ensures both causal reliability and algorithmic feasibility. Through extensive verification, we find that (i) the identified paths are the primary sources for decision signals, as evidenced by a marked decrease in self-repair compared to non-path components; and (ii) transformers utilize modular, class-shared path mechanisms to process categorical information. Beyond these findings, we show that the potential implications of our method can stem from sparse manipulation, highlighting efficient applicability to downstream tasks such as model debugging and pruning.

Large Reasoning Models (LRMs) exhibit backtracking and self-verification mechanisms that enable them to revise intermediate steps and reach correct solutions, yielding strong performance on complex logical benchmarks. We hypothesize that such behaviors are beneficial only when the model has sufficiently strong “critique” ability to detect its own mistakes. This work systematically investigates how current LRMs recover from errors by inserting arithmetic mistakes in their intermediate reasoning steps. Notably, we discover a peculiar yet important phenomenon: despite the error propagating throughout the entire chain-of-thought (CoT) without any verbalized correction, the model still reaches the correct final answer after the thinking process finishes. This recovery implies the existence of an internal mechanism helping the model to detect errors and trigger self-correction, which we refer to as the hidden critique ability. Building on feature space analysis, we identify a highly interpretable critique vector representing this behavior. Extensive experiments across multiple model scales and families demonstrate that steering latent representations with this vector improves the model’s error detection capability and enhances the performance of test-time scaling at no extra training cost. Our findings provide a valuable understanding of LRMs’ critique behavior, suggesting a promising direction to control and improve their self-verification mechanism.

Current text-to-3D methods excel at generating single objects but falter on compositional prompts. We argue this failure is fundamental to their optimization schedules, as simultaneous or iterative heuristics predictably collapse under a combinatorial explosion of conflicting gradients, leading to entangled geometry or catastrophic divergence. In this paper, we reframe the core challenge of compositional generation as one of optimization scheduling. We introduce DecompDreamer, a framework built on a novel staged optimization strategy that functions as an implicit curriculum. Our method first establishes a coherent structural scaffold by prioritizing inter-object relationships before shifting to the high-fidelity refinement of individual components. This temporal decoupling of competing objectives provides a robust solution to gradient conflict. Qualitative and quantitative evaluations on diverse compositional prompts demonstrate that DecompDreamer outperforms state-of-the-art methods in fidelity, disentanglement, and spatial coherence.

Single-cell multi-omics technologies jointly measure multiple modalities from the same cell, such as gene expression, chromatin accessibility, and surface protein abundance, providing a richer view of cellular state than any single modality alone. A central challenge is how to learn a multimodal cell representation that preserves the different types of information contained in these co-profiled measurements. Existing integration methods often emphasize either cross-modal correspondence, which benefits retrieval and imputation, or fused discriminative structure, which benefits classification, but these objectives can favor different aspects of the data. As a result, a representation optimized for one setting may fail to preserve information needed for another. We introduce scDecomp, a decomposed representation learning framework for co-profiled single-cell multi-omics data. Motivated by the concepts of redundancy, uniqueness, and synergy, scDecomp separates multimodal information into three role-specific components: a redundancy component ($R$) for information shared across modalities, a uniqueness component ($U$) for modality-specific information, and a synergy component ($S$) for interaction-dependent multi-omic signals. This design allows the learned representation to retain shared cell-state structure while also preserving modality-specific and cross-modal regulatory information. On single-cell multi-omics benchmarks, the $R$ component achieves a 5.0\% reduction in FOSCTTM compared with the strongest baseline, while $U+S$ improves cell type classification accuracy by 1.38\%. Branch-level analyses further show that the decomposed components support complementary aspects of representation quality, suggesting that structured decomposition provides a practical strategy for learning more informative multimodal cell representations.

Embodied navigation and reasoning require an agent to actively acquire, organize, and verify task-relevant evidence in previously unseen environments. Recent methods typically rely on semantic maps, scene graphs, vision-language priors, or structured memory, but often underuse persistent fine-grained 3D evidence and entangle query-related ambiguity with environment-side reliability. This coupling can make evidence acquisition inefficient and final decisions vulnerable to weak or transient observations. We propose Decoupled Complementary Fields on 3D Gaussian Maps, an embodied navigation framework that maintains persistent volumetric evidence, separates query-related ambiguity from environment-side readiness, and grounds decisions through progressive verification. First, we construct a query-conditioned object-centric Gaussian navigation state built on online 3D Gaussian Splatting, which preserves fine-grained 3D evidence across viewpoints while providing an executable spatial interface for region reasoning and navigation. We further formulate Decoupled Complementary Fields, consisting of a Query-Ambiguity Field, which captures where task-relevant evidence remains unresolved, and an Environment-Readiness Field, which estimates where the environment is structurally reliable for motion and inspection. Finally, we couple these fields in a hierarchical navigation-and-verification policy, which selects informative frontiers, instantiates executable local goals, and verifies task-relevant candidates using persistent volumetric Gaussian evidence before stopping or answer handoff. Experiments on A-EQA and GOAT-Bench show consistent gains in active evidence acquisition, multi-modal lifelong navigation, and reliable final decision making.

Recent full-duplex spoken dialogue models have demonstrated compelling progress toward human-like interaction, enabling agents to respond with low latency, produce backchannels, and handle user barge-ins. Yet these improvements in conversational dynamics often come with weaker reasoning and instruction-following abilities, revealing a potential tension between interactive dynamics and intelligence capability. In this paper, we argue that such an intelligence--dynamics trade-off is not fundamental: conversational dynamics can instead be learned as a separate real-time decision policy from human dialogue data. To this end, we propose DuplexPO, a reinforcement learning (RL) framework that decouples when to speak from what to say. It preserves the semantic response capability of an instruction-tuned assistant, while optimizing its temporal interaction behavior over selected high-impact windows from long human conversations. To quantitatively optimize these dynamics, we formulate the Factorized Conversational Dynamics Reward (FCDR) to enable fine-grained temporal credit assignment for turn initiation, backchanneling, yielding, and regularized participation. The policy is then optimized with a GRPO-style objective. Experiments show that DuplexPO substantially improves full-duplex behaviors, including timely backchannels, smooth turn-taking, and barge-in handling, while maintaining strong reasoning and instruction-following performance. Moreover, improvements in dynamics-oriented metrics are reflected in better user experience, suggesting that optimizing conversational timing as a standalone objective can promote more natural full-duplex interaction.

Distributional reinforcement learning (RL) is a powerful framework increasingly adopted in safety-critical domains for its ability to optimize risk-sensitive objectives. However, the role of the discount factor is often overlooked, as it is typically treated as a fixed parameter of the Markov decision process or tunable hyperparameter, with little consideration of its effect on the learned policy. In the literature, it is well-known that the discounting function plays a major role in characterizing time preferences of an agent, which an exponential discount factor cannot fully capture. Building on this insight, we propose a novel framework that supports flexible discounting of future rewards and optimization of risk measures in distributional RL. We provide a technical analysis of the optimality of our algorithms, show that our multi-horizon extension fixes issues raised with existing methodologies, and validate the robustness of our methods through extensive experiments. Our results highlight that discounting is a cornerstone in decision-making problems for capturing more expressive temporal and risk preferences profiles, with potential implications for real-world safety-critical applications.

Double Q-learning is a classical control algorithm that mitigates the maximization bias of Q-learning. To do so, it explicitly trains two independent action-value functions and uses them to decouple action-selection and action-evaluation when computing bootstrap targets. Double DQN adapts target bootstrap decoupling to deep reinforcement learning (RL), but explicitly trains only a single action-value function and does not fully decouple its estimators. Consequently, the two estimators remain correlated, and overestimation persists. In this paper, we introduce Deep Double Q-learning (DDQL), a deep RL algorithm that explicitly trains two Q-functions through Double Q-learning. DDQL stabilizes training through a combination of techniques, including lower replay ratios, longer target network update intervals, and shared layers. Across 57 Atari 2600 games, DDQL improves aggregate performance over Double DQN, outperforming it on 47 games while further reducing overestimation. In addition, we study key design choices when adapting Double Q-learning to deep RL, including the network architecture, replay ratio, and minibatch sampling strategies.

Phylogenetic inference plays a central role in understanding evolutionary relationships, with applications ranging from tracking pathogen spread to reconstructing the history of life. Conventionally, practitioners obtain posteriors over the phylogenetic tree of species using MCMC methods. However, such likelihood-based methods are only tractable for simple evolutionary models with restrictive assumptions. For more complex and realistic evolutionary models, conventional methods are prohibitively expensive, inaccurate, or even impossible. We instead advocate for a simulation-based inference approach, by using simulated data from an evolutionary model to train a neural network that predicts tree topologies conditioned on sequences. To accurately represent the complex posterior distributions over tree topologies that can arise, we present flexible models that iteratively generate trees using three natural paradigms: top-down, middle-out, and bottom-up. We use a discrete diffusion framework to train these models efficiently on large-scale simulated datasets of phylogenetic trees. For all three generative paradigms, our models fit the data substantially better than the previous state-of-the-art simulation-based method, Phyloformer 2, and obtain more accurate posteriors on real datasets. Finally, our models outperform misspecified conventional methods on data following complex evolutionary processes.

Neural representations are not unique objects. Even when two systems realize the same downstream computation, their hidden coordinates may differ by reparameterization. A probe family intended to reveal structure already present in a representation should therefore be stable under the relevant representation symmetries rather than be tied to a particular basis. We study this group action in the tractable exact setting of the final readout layer, where equivalent realizations induce affine changes of hidden coordinates. The resulting symmetry principle singles out a unique hierarchy of shallow coordinate-stable probes, with linear probes as its degree-1 member. We also show that a natural object for cross-model probe transfer is a shared probe-visible quotient--the representation modulo directions invisible to the probe family--rather than the full hidden state. Experiments on synthetic and real-world tasks support both predictions, showing where degree-2 probes help beyond linear ones and how quotient-based transfer enables coverage-aware monitor portability across model families. These results point toward a broader geometric representation theory of neural probing, with coverage-aware monitor transfer as a concrete operational consequence.

We present DeFlow, a decoupled offline RL framework for extracting high-value actions from a learned multi-step flow prior. Directly optimizing iterative generative policies typically requires backpropagation through ODE solvers, while shortcut policies trade off expressivity, inference cost, and policy-improvement objectives. DeFlow keeps the flow model as a behavior prior and trains a lightweight action-conditioned residual module for value improvement under an adaptive trust-region penalty. This design bypasses solver differentiation and encourages proximity to the learned prior without treating the trust region as exact data-support preservation. Empirically, DeFlow is competitive with recent generative offline RL baselines, with the clearest gains on selected multimodal manipulation and offline-to-online settings.

Imitation learning has achieved remarkable progress in robot manipulation, but its success heavily depends on large-scale demonstration data that are costly to collect. Recent work has therefore explored demonstration augmentation, but they are fundamentally limited in deformable manipulation, where task-relevant object variation is governed by high-dimensional deformations and physics-induced internal constraints rather than low-dimensional pose changes. We present DeformGen, a dynamics-based augmentation framework that achieves topological diversity for deformable objects. Instead of perturbing object pose, DeformGen expands the valid state distribution by applying localized physical disturbances, forward-simulating the dynamics, and stabilizing the result to obtain topology-coherent deformable states with physical plausibility. Given these synthesized states, DeformGen further transfers source manipulation trajectories via deformation-field warping, which lifts per-particle displacements into a continuous spatial function to adapt the end-effector trajectory consistently with the deformed geometry. In this way, our method augments both the state distribution and its associated manipulation behavior. Experiments on high-fidelity deformable manipulation benchmarks show that DeformGen consistently improves policy learning over training on the original demonstrations alone and over rigid-style augmentation baselines.

Recent feed-forward 3D reconstruction transformers have scaled to over a billion parameters, following the broader trend of increasing model capacity in computer vision. Yet emerging evidence suggests that contiguous transformer layers often behave like repeated applications of similar operations, and multi-view reconstruction transformers refine their predictions progressively across decoder depth. We posit that model depth partially buys iteration, paid for inefficiently in unique parameters, and instead make that iteration explicit in architecture. Our model, DéjàView, applies a single looped transformer block recurrently to per-view features for K refinement steps. Trained once, it exposes K as an inference-time compute knob, matching or outperforming substantially larger feed-forward baselines across five reconstruction benchmarks spanning indoor, outdoor, object-centric, and driving scenes, while using a fraction of their parameters and comparable or lower compute. Importantly, the same looped block formulation outperforms an otherwise identical variant with independent per-step parameters under matched training data and compute, suggesting that explicit iteration is not merely a compute-efficient substitute for capacity but a stronger inductive bias for multi-view 3D reconstruction.

Distributed reinforcement learning trains on data from stale, buggy, or mismatched actors, producing actions with high surprisal (negative log-probability) under the learner's policy. The core difficulty is not surprising data per se, but \emph{negative learning from surprising data}. High-surprisal failures can dominate finite-batch updates through large perpendicular components, while high-surprisal successes reveal opportunities the current policy would otherwise miss. The \textit{Delightful Policy Gradient} (DG) separates these cases by gating each update with delight, the product of advantage and surprisal, suppressing rare failures and preserving rare successes without behavior probabilities. In a tabular analysis, DG suppresses the perpendicular second moment of high-surprisal failures by a policy-overlap factor that vanishes as the learner improves. The advantage sign is essential for surprisal-based filtering: any learner-probability-only gate that suppresses rare failures also suppresses rare successes. On MNIST with simulated staleness, DG without off-policy correction outperforms importance-weighted PG with exact behavior probabilities. On a transformer sequence task with staleness, actor bugs, reward corruption, and rare discovery, DG often achieves nearly order-of-magnitude lower error. When all four frictions act simultaneously, its sample-efficiency advantage is order-of-magnitude and grows with task complexity.

Autoregressive (AR) text-to-speech (TTS) models generate speech tokens one at a time, and inference latency therefore scales linearly with output length. Discrete diffusion language models (dLLMs) have recently emerged as a parallel alternative that produces tokens via iterative unmasking. Recent work has converted pretrained AR language models into dLLMs in text generation, but this paradigm has not been extended to speech synthesis. In addition, existing conversion methods require full fine-tuning and large-scale training data. This leaves the question of whether such conversion can be done with substantially less compute and data largely open. We introduce DELTA-TTS, a lightweight conversion that turns a pretrained AR TTS backbone into a dLLM. The AR weights are kept frozen; all adaptation is routed through LoRA and a per-block speech-aware convolution. The convolution injects local acoustic context into the bidirectional attention, supplying the short-range continuity between adjacent speech tokens. With only $585$ hours of LibriTTS as adaptation data, DELTA-TTS achieves a state-of-the-art WER of $\textbf{1.75}\%$ on Seed-TTS test-en and decodes $\textbf{3.3}\times$ faster on the token-generation stage than the CosyVoice3 AR backbone it is converted from. This shows that lightweight AR-to-dLLM conversion provides a practical, data- and compute-efficient route to non-autoregressive TTS.

While monocular depth estimation has achieved significant progress, achieving generalized metric depth estimation for both narrow field-of-view (FoV) perspectives and $360^\circ$ panoramas remains an unsolved challenge. Existing methods are often tailored to specific camera types and struggle to produce accurate metric depth that generalizes across diverse settings. This limitation stems from two key challenges: the inherent geometric discrepancy between perspective and panoramic cameras, and the scarcity of panoramic training data with metric annotations. In this work, we introduce \textbf{DepthMaster}, a unified metric depth estimation framework. Rather than employing specialized networks to learn spherical distortions, we reformulate the problem by decomposing panoramic images into overlapping perspective patches. Crucially, distinct from prior projection-based methods that rely on ad-hoc architectural modifications to handle boundaries, we introduce a novel Correspondence Consistency Loss (CCL) and inject virtual projection cameras as geometric priors, allowing us to seamlessly stitch the patches while avoiding specialized operators and keeping the backbone largely compatible with standard Transformer designs. This strategy also resolves the geometric differences by unifying all inputs into a canonical perspective representation, and effectively circumvents data scarcity by directly unlocking powerful metric priors from vast perspective datasets. Trained on a mixed dataset that contains only one panorama dataset, DepthMaster achieves state-of-the-art zero-shot performance on 13 diverse datasets, outperforming not only universal methods but also leading specialist models in both perspective and panoramic domains. The code and models of DepthMaster will be released.

Modern automatic speech recognition (ASR) systems have achieved remarkable accuracy by scaling model depth and capacity, but at the cost of substantial memory and computation. On edge devices where ASR is often most needed, such as watches and glasses, deploying such large models is infeasible due to extremely constrained resources. This raises a fundamental question: can we achieve high representational power in deep ASR models without scaling up parameterization? In this work, we revisit the role of depth and identify layer-wise representational dynamics, in which most layers learn functionally similar transformations. Motivated by this insight, we propose a block-recurrent ASR architecture that replaces parameterized depth with a small set of recurrent blocks, each consisting of weight-shared layers, thereby preserving effective depth while drastically reducing model size. Through representation-guided grouping and knowledge distillation, block-recurrent models retain the accuracy of large reference models while using an order of magnitude fewer parameters. Across Open ASR Leaderboard benchmarks, a model with only two shared blocks recovers $97.4$% of the accuracy of large models on average. The compact backbone further enables efficient on-device personalization through lightweight MoE-LoRA adaptation, allowing user-specific models to match or surpass the accuracy of general-purpose ASR systems with significantly lower inference cost. Together, these results show that high-quality ASR can be achieved without large parameterization, and that structured recurrence provides a principled path towards ultra-compressed, high-efficiency, and personalized speech recognition on the edge.

Out-of-domain (OOD) robustness is challenging to achieve in real-world computer vision, especially in unsupervised domain adaptation scenarios, where shifts in image background, style, and acquisition instruments often degrade model performance. Generic augmentations show inconsistent gains under such shifts, whereas dataset-specific augmentations require expert knowledge and prior analysis. Moreover, prior studies show that neural networks adapt poorly to domain shifts because they exhibit a learning bias to domain-specific frequency components. Perturbing frequency values can mitigate such bias but overlooks pixel-level details, leading to suboptimal performance. To address these limitations, we propose D-GAP, a Dataset-agnostic and Gradient-guided augmentation method for the Amplitude spectrum (in frequency space) and the Pixel values. Unlike conventional handcrafted augmentations, D-GAP computes sensitivity maps in the frequency space from task gradients, which reflect how strongly the deep models respond to different frequency components, and uses the maps to adaptively interpolate amplitudes between source and target samples. We further propose a dual-space augmentation that jointly controls spectral bias and spatial fidelity by introducing a complementary pixel-space blending branch. This way, D-GAP turns augmentation from fixed, random, or manually designed perturbation into a model-response-adaptive intervention. Extensive experimental results show that the proposed method consistently outperforms both generic and dataset-specific domain adaptation methods, improving average OOD performance by +5.3% on four real-world datasets and +1.9% on three benchmark datasets.

Evaluating LLM-generated interactive software requires execution in addition to static analysis. The key difficulty is that correctness is a graph-level reachability property over latent UI state-transition graphs, whereas a GUI evaluator observes only a single execution trajectory. A failed rollout therefore rules out only one realized path, leaving failure attribution ambiguous between evaluator-side execution error and genuine software defect. We present DIAGEVAL, a trajectory-conditioned diagnostic evaluation protocol for post-failure GUI-agent evaluation of interactive software. Rather than blindly retrying from scratch,DIAGEVAL reuses the failed trajectory to choose targeted diagnostic probes and aggregates their outcomes into an internal attribution signal. The latent-graph view motivates the diagnostic problem; DIAGEVAL does not reconstruct the graph or estimate calibrated posterior probabilities. We evaluate DIAGEVALon WebDevJudge-Unit and RealDevBench across multiple GUI-agent evaluators and LLM backbones. On false-negative cases, DIAGEVAL recovers 45.6--62.1\% of failures that were initially attributed to software defects, outperforming retry-based baselines with 34.4--160.6\% relative gains. On the full evaluation sets, this recovery improves accuracy from 69.9\% to 78.3\% on WebDevJudge-Unit and from 65.0\% to 81.6\% on RealDevBench. These results suggest that reliable GUI-agent evaluation requires not only stronger execution, but also active failure diagnosis to disambiguate evaluator-side errors from genuine software defects.

Long-video question answering is challenging since the answer often hinges on a few decisive moments scattered throughout the video, while memory constraints force Multimodal Large Language Models (MLLMs) to sample frames at extremely low rates, resulting in extreme sparsity that risks omitting key evidence. Though aggressively enlarging the context size is intuitive, we argue that performance remains fundamentally constrained by the bias, $\textit{i.e.}$, the misleading cues caused by spurious linguistic correlations and salient yet irrelevant visual observations. In this paper, we first introduce S-MME, a benchmark designed to systematically diagnose shortcut-biased behavior in MLLMs. To mitigate this failure, we further propose a training-free framework, Counterfactual Long-video Evidence-Aware Reasoning (CLEAR). By treating the full video-question pair as a complete view, CLEAR estimates bias effects via other views by only keeping the question or locally salient visual clips. Then, we apply hidden-state intervention to mitigate the discrepancy at inference such that the correct answer can be predicted confidently. With experimental studies across multiple benchmarks, we show that our method is generalizable and can be integrated in different MLLMs, yielding consistent gains in prediction and robustness. We hope CLEAR can offer a complementary direction in test-time scaling for long-video understanding and will publish code.

Multimodal Large Language Models demonstrate strong performance on multimodal benchmarks, yet remain fragile when one modality carries spurious or misleading content. We trace this fragility to a deficiency in cross-modality competency, defined as the ability to fairly evaluate and integrate information across modalities, and identify its concrete, measurable manifestation as \emph{Modality Interference}, in which task-irrelevant modality signals improperly influence the model's predictions. To diagnose this phenomenon systematically, we design a perturbation-based evaluation grounded in causal intervention, in which controlled noise is injected into the task-irrelevant modality across image-heavy and text-heavy tasks. Across diverse MLLM families and scales, we observe consistent performance degradation on modality-heavy tasks under perturbations, indicating that modality interference is a pervasive and scale-resistant failure mode rather than an artifact of any specific architecture. To mitigate this, we propose a unified perturbation-aware fine-tuning framework that combines (i) heuristic and adversarial data augmentation targeting the task-irrelevant modality, and (ii) output-level consistency regularization between clean and perturbed inputs. Extensive experiments across diverse MLLM architectures, model scales, and benchmarks show that our approach simultaneously improves unimodal robustness and standard multimodal performance, achieving Pareto-optimal gains over existing baselines.

Monitoring complex industrial assets relies on engineer-authored symbolic rules that trigger on sensor conditions and prompt technicians to perform corrective actions. The bottleneck is not detection but response: translating rules into maintenance steps requires asset-specific knowledge gained through years of practice. We investigate whether LLMs can serve as decision support for this rule-to-action step and introduce DiagnosticIQ, a benchmark of 6690 expert-validated multiple-choice questions from 118 rule-action pairs across 16 asset types. We contribute (i) a symbolic-to-MCQA pipeline normalizing rules to Disjunctive Normal Form with embedding-based distractor sampling, (ii) five variants probing distinct failure modes (Pro, Pert, Verbose, Aug, Rationale), and (iii) a benchmark of 29 LLMs and 4 embedding baselines. A human evaluation (9 practitioners, mean 45.0\%) confirms DiagnosticIQ requires specialist knowledge beyond operational experience. Three findings stand out. The frontier has closed: the top three LLMs lie within one Macro point, with Bradley-Terry Elo placing claude-opus-4-6 30 points above the next model. Yet DiagnosticIQ-Pro exposes brittleness, with every model losing 13-60\% relative accuracy under distractor expansion. DiagnosticIQ-Aug exposes pattern-matching: under condition inversion, frontier models still select the original answer 49-63\% of the time. The deployment bottleneck is not capability but calibration: frontier models handle template-style fault detection but break under structural perturbation.

Cross-entropy (CE) loss is central to deep learning, but existing theory often relies on simplifications—such as squared loss or convex models—that miss key aspects of CE optimization. In this work, we study multi-class CE dynamics using a two-layer linear network with orthogonal inputs, the simplest non-convex setting where the CE implicit bias remains unresolved. This coincides with the unconstrained features model used to study neural collapse (NC). Our analysis is based on a key observation: Hadamard initialization diagonalizes the softmax operator. This allows us to extend the spectral initialization framework that Saxe et al. (2013, 2019) developed for squared loss. We prove convergence to NC under spectral CE training and give the first finite-time analysis in this setting via an explicit Lyapunov function that decreases monotonically to NC despite spurious critical points. We further identify CE-specific phenomena absent under squared loss—coupling, non-monotonic convergence, qualitative dependence on the number of classes, and exponential slowdowns. We also characterize the role of network width and show empirically that spectral dynamics qualitatively model small random initialization.

Recent advances in 3D Gaussian Splatting (3DGS) present two main directions: feed-forward models offer fast inference in sparse-view settings, while per-scene optimization yields high-quality renderings but is computationally expensive. To combine the benefits of both, we introduce Diff3R, a novel framework that explicitly bridges feed-forward prediction and test-time optimization. By incorporating a differentiable 3DGS optimization layer directly into the training loop, our network learns to predict an optimal initialization for test-time optimization rather than a conventional zero-shot result. To overcome the computational cost of backpropagating through the optimization steps, we propose computing gradients via the Implicit Function Theorem and a scalable, matrix-free PCG solver tailored for 3DGS optimization. Additionally, we incorporate a data-driven uncertainty model into the optimization process by adaptively controlling how much the parameters are allowed to change during optimization. This approach effectively mitigates overfitting in under-constrained regions and increases robustness against input outliers. Since our proposed optimization layer is model-agnostic, we show that it can be seamlessly integrated into existing feed-forward 3DGS architectures for both pose-given and pose-free methods, providing improvements for test-time optimization.

Diffusion models have become a widely used framework for probabilistic time series forecasting, modeling the distribution of future values given an observed history. In time series forecasting, however, the future continues the observed history, creating an asymmetry the standard diffusion process leaves unaddressed, with slowly-varying content largely determined by the observed continuity while higher-frequency dynamics carry most of the residual uncertainty. Existing diffusion-based forecasters decouple this asymmetry through an external rule before generation, leaving the corruption trajectory blind to which parts of the target the history can already anchor. We propose \textsc{DiffDiff}, a diffusion framework that embeds this predictability asymmetry into the diffusion trajectory itself, so that a single end-to-end diffusion process becomes aware of which parts of the target the history can already anchor. \textsc{DiffDiff} makes the forward operator step-dependent so that the noisy intermediate state progressively shifts from the target itself toward its second-order differenced structure, while a conditioning pathway supplies the denoiser with both value-domain and differential history information balanced by a stage-adaptive gate at each diffusion step. The terminal distribution approaches a standard Gaussian, preserving compatibility with existing samplers. On seven benchmarks across four prediction horizons, \textsc{DiffDiff} outperforms six diffusion baselines, and our analysis confirms that \textsc{DiffDiff} concentrates the diffusion's generative effort on the most uncertain components of the target while relieving it from rebuilding the history-anchored content.

Predicting transcriptional responses to genetic perturbations is fundamental to functional genomics and therapeutic discovery. Recent deep learning models have shown promise in single-cell perturbation response prediction, but they typically generate each response in isolation, without explicitly leveraging experimentally characterised responses to related perturbations. We introduce PT-RAG (Perturbation-aware Two-stage Retrieval-Augmented Generation), a plug-in retrieval-and-conditioning module for generative cellular perturbation response. PT-RAG augments an existing perturbation-response backbone with learned access to related perturbation contexts. The key challenge is that relevance is not fixed in this setting: functionally related genes may elicit different effects across cell types. PT-RAG addresses this with a two-stage retrieval mechanism: GenePT-based semantic retrieval first identifies $K$ candidate perturbations, after which a differentiable Gumbel-Softmax selector adaptively selects retrieved contexts conditioned on the control cell state, the query perturbation, and each candidate perturbation. Through cross-cell-type and cross-perturbation generalization tasks, PT-RAG consistently improves distributional similarity and often overall predictive quality; for example, on scGPT cross-cell-type results, Wasserstein distance drops by 10.7%. The code to reproduce our experiments is available at https://anonymous.4open.science/r/PT-RAG_NIPS-13D8/.

In machine learning applications, privacy requirements during inference or deployment time could change constantly due to varying policies, regulations, or user experience. In this work, we aim to generate a magnitude of models to satisfy any target differential privacy (DP) requirement without additional training steps, given a set of existing models trained on the same dataset with different privacy/utility tradeoffs. We propose two post processing techniques, namely random selection and linear combination, to output final private models satisfying any target privacy parameter. We provide privacy accounting of these approaches from the lens of R\'enyi DP and privacy loss distributions for general problems. In a case study on private mean estimation, we precisely characterize the privacy/utility results and theoretically establish the superiority of linear combination over random selection. Empirically, we validate our approaches and analyses on several models and both synthetic and real-world datasets.

Traditional neural solvers for solving vehicle routing problems (VRPs) often suffer from limited solution diversity, motivating the development of Generative Flow Network (GFlowNet)–based models. However, the effectiveness of these models is frequently constrained by insufficient flow expansion in high-reward regions, limiting their ability to distribute probability across promising solution routes, the deeper exploration could yield superior results. Diffusion models, in contrast, provide stronger structural guidance for exploration. These two paradigms are naturally complementary: GFlowNet can supply edge-level signals for diffusion to embed, while diffusion can guide broader exploration. Leveraging this synergy, we propose Diffusion-Enhanced GFlowNet (DEG), a novel framework that integrates GFlowNet with diffusion model to encourage richer flow expansion toward high-reward regions and derive higher-quality solutions. Specifically, DEG exploits GFlowNet’s inherent diversity to generate edge-specific backward signals, applies the stochastic noise schedule of diffusion to perturb these signals, and then denoises them within the GFlowNet paradigm. To further improve scalability, we introduce a specialized decoder capable of dynamically adapting to diverse problem scales. Extensive experimental evaluations on synthetic and real-world datasets, including instances with up to 10,000 nodes, demonstrate that DEG consistently achieves favorable performance compared to baseline methods.

Prompt learning is a popular parameter-efficient method for adapting foundation models, but learned prompts are typically task-specific and fail to generalize to new classes, domains, or compositions of tasks. In this paper, we introduce a $\textbf{Diffusion Meta-Prompt (DMP)}$ model , a framework that models the distribution of learned prompts using diffusion models. DMP is trained only on a repository of previously learned prompts to synthesize new prompts conditioned on natural language task descriptions, without access to the task data. To improve the sampling stability, we introduce a test-time steering strategy for DMP, which uses the best training-selected prompt in the repository as a latent anchor during diffusion sampling, without retraining the DMP or accessing test classes. DMP improves generalization across classification, retrieval and text-to-image generation tasks, supports concept composition and negative prompting without explicit training. It reduces storage and inference costs by over 90% compared to prompt retrieval methods. For composite classification, DMP achieves upto $\textbf{2.0}$% average gain over prior meta-learning methods across 55 pairs of datasets with gains as high as $\textbf{8.5}$% on specific pairs such as Eurosat and Flowers. DMP also enhances cross-task generalization with $\sim$$\textbf{2-9}$% improvement for hierarchical classification task. We further provide a theoretical guarantee bounding the expected task loss of prompts sampled from a DMP.

Classifier-free Guidance (CFG) is central to modern diffusion models, but operates as an inference-time correction where the training objective learns the unbiased conditional score, while guided sampling targets a posterior-tilted score that never explicitly learned. Rather than another inference-time patch or distillation, Model-guidance (MG) directly integrates the posterior tilt into the training objective and eliminates CFG during inference. MG serves as a plug-and-play module compatible with existing methods, yet accelerates convergence by online self-bootstrapping and doubles inference speed. Unlike distillation surrogates, a reparameterisation property reveals that MG implicitly learns the vanilla score without mode collapse. Extensive experiments demonstrate that MG matches and even surpasses CFG baselines, and achieves a state-of-the-art FID of $1.34$ on ImageNet $256$ benchmark.

Long-form spatial reasoning in vision--language models (VLMs) is often bottlenecked by autoregressive chain-of-thought (CoT) decoding, where intermediate reasoning traces and final answers are generated token by token. This sequential decoding path makes high-quality spatial reasoning expensive in latency-sensitive embodied and interactive settings. We propose \emph{Diffusion Thinking}, a post-training framework that converts pretrained autoregressive VLMs into diffusion-style reasoners for fast long-form spatial inference without changing the visual encoder or language backbone. Diffusion Thinking partitions each CoT trace into blocks and refines tokens within the current block in parallel through discrete denoising, while preserving causal dependence across blocks and reusing KV cache for streaming generation. This creates a high-speed operating point, but aggressive parallel refinement can weaken fine-grained reasoning. To address this, we introduce \emph{Diffusion Rethinking}, a cached test-time scaling mechanism that reallocates part of the saved latency budget to self-correction rounds, yielding a controllable speed--accuracy frontier. We construct a 561K-instance Long-CoT Spatial Reasoning dataset with verified final answers and generated CoT traces across six spatial task types. Across Qwen2.5-VL and InternVL3 backbones, Diffusion Thinking with block size $D=32$ achieves 38.18--41.81$\times$ effective wall-clock speedup over autoregressive Long-CoT decoding. With eight rethinking rounds, it retains 4.20--4.60$\times$ speedup and matches or surpasses autoregressive Long-CoT accuracy on the largest backbones. These results show that diffusion-style compute reallocation is a practical path toward fast, accurate long-form spatial reasoning in VLMs.

Open-vocabulary 3D Gaussian segmentation typically binds semantic features to per-Gaussian state via per-scene gradient training (1–4 h) or the more recent closed-form distillation. Both commit each primitive to a single feature; therefore, boundary Gaussians spanning multiple 2D instances are forced into a hard decision that degrades retrieval accuracy and yields unstable mask boundaries at render time. We present DIGS, a training-free framework with two stages, class-agnostic map- ping and multi-modal encoding, that decouples instance discovery from linguistic naming. At the Gaussian level, our distribution-informed representation maintains an explicit top-M label distribution per primitive, updated by a tempered Dirichlet rule; preservation of multi-peaked posteriors at boundary primitives allows a late-argmax renderer to recover clean instance boundaries. A bidirectional state machine aggregates these distributions into a persistent class-agnostic instance ontology. The instance-level encoding stage then fuses the CLIP visual embedding of each instance’s isolated 3D footprint with the CLIP-text embedding of an LLM-generated attribute description, disambiguating visually similar instances. DIGS maps a scene in 1–3 minutes on a single A100 and serves 3D-native retrieval in 5.4 ms, reaching 71.95% mIoU on LERF-OVS, surpassing the strongest training-free baseline SFS and the strongest training-based baseline LaGa, and setting new state-of-the-art performance on LERF-Mask (91.71%) and 3D-OVS(96.45%). Codes are at: https://anonymous.4open.science/r/DIGS-C4EF.

Towards more general and human-like intelligence, large language models should seamlessly integrate both multilingual and multimodal capabilities; however, extending an existing multimodal model to many languages typically requires expensive multilingual multimodal data construction and repeated end-to-end retraining. We study a training-free alternative: injecting multilingual capability into an existing multimodal model by composing residual updates in the shared language model backbone. The key challenge is that multilingual and multimodal updates are heterogeneous, reflecting different functional roles in the shared model. To address this, we propose Direction- and Magnitude-aware Multilingual Multimodal merging (DiM$^3$), which selectively composes the two updates at each parameter dimension while preserving the original vision encoder and multimodal projector. Experiments on multilingual benchmarks in both text-only and vision-language settings, covering 57 languages across LLaVA- and Qwen-based backbones, show that DiM$^3$ consistently outperforms existing merging baselines, substantially improves multilingual performance over the original multimodal model, and remains competitive with dedicated multilingual multimodal fine-tuning while largely retaining general multimodal ability. We further show that DiM$^3$ can be directly applied to already trained multilingual multimodal models and still yield additional gains. Further interpretability analysis shows that DiM$^3$ primarily reshapes intermediate-layer semantic representations, strengthening cross-lingual alignment under both text-only and multimodal inputs while preserving higher-layer task-sensitive structure. Our repository is on https://anonymous.4open.science/r/w-C677/.

Diffusion models are a leading paradigm for graph generation, with notable impact in domains such as molecular design. Yet, scaling these models to large graphs remains an open problem. We approach this question in the dense-graph setting through the lens of graphons, the size-agnostic limit object of dense graph sequences, to study how structural graph statistics behave across node-size scales. This perspective leads to DiPhon, a diffusion process for size-scalable graph generation. Specifically, we formulate a continuous diffusion process on the graphon space via a Jacobi stochastic differential equation (SDE), and propose DiPhon, a discretized scheme that imitates it on graphs. We further derive the corresponding reverse-time process, which requires access to the marginal score. For the Jacobi process, this score interestingly admits a tractable form, which we estimate from data via graph denoising and plug into the reverse process to generate graph samples. We prove that DiPhon matches the first moment of the marginal distributions induced by the continuous graphon process exactly, and approximates the second moment up to a closed-form discrepancy. In this way, DiPhon inherits the size-agnostic statistical properties of the graphon dynamics can be used to solve the scaling bottleneck. Empirically, we demonstrate the scalability of DiPhon by training on small graphs and generating substantially larger ones at inference time, without retraining on large graphs.

LLM-based coding agents have shown strong capabilities on real-world software testing tasks, yet existing evaluations concentrate on unit test generation, leaving system testing unexplored. System testing is both practically important for software quality assurance and inherently difficult, as it demands repository-level holistic understanding and long-horizon reasoning over program behavior. In this paper, we propose to measure LLM agents' system testing capabilities through its core problem, Directed Input Synthesis (DIS): generating system-level inputs that exercise specified code branches. We introduce DIS-Bench, a benchmark built via an automated pipeline based on random testing that selects target branches that are both reachable and challenging. Because DIS-Bench is constructed from execution behavior rather than human-authored issues or patches, it is scalable and less susceptible to contamination from existing training corpora. DIS-Bench contains 8267 tasks drawn from 37 real-world repositories. We evaluate 7 mainstream open-sourced LLMs on DIS-Bench-Lite under both BM25 retrieval and state-of-the-art coding agents, including OpenHands, Mini-SWE-Agent, and Claude-Code. The results show that DIS is highly challenging for current models: the best-performing configuration resolves only 28.1\% of targets at pass@1. We further perform a manual inspection to characterize the dominant failure modes, revealing reasoning bottlenecks of LLMs on repository-level system testing tasks.

Cells in multicellular systems have complex shapes, yet most techniques for predicting biological dynamics from image sequences discard shape information by reducing cells to point particles. Here, we present a framework that moves beyond point representations by learning dynamic, shape-aware representations of synthetic cells in a multicellular collective directly from 2D time-series images. Specifically, our GraphDINO framework combines a frozen DINOv3 backbone for shape representation with a graph neural network trained through next-frame prediction. GraphDINO learns a per-cell latent representation that captures underlying cellular properties such as growth rate, membrane tension, and cell-cell adhesion without parameter supervision. The quality of this recovery depends on the encoder, with DINOv3 features providing a better shape representation than classical shape encoders. Together, these results demonstrate that combining foundation model features with graph-based interaction rules allows the discovery of key dynamical parameters that govern cellular behavior, directly from image sequences.

We study posterior sampling for inverse problems in discrete state spaces using discrete diffusion models as generative priors. While continuous diffusion models have become widely used for inverse problems, their discrete counterparts remain comparatively underexplored. Existing discrete posterior samplers often rely on continuous relaxations of discrete variables, Gibbs-style updates, or mechanisms specialized to particular corruption processes, which can limit scalability or generality. We propose $\Delta$LPS, a Discrete Langevin-Inspired Posterior Sampler that uses gradient information to identify promising discrete moves without leaving the discrete state space. The resulting approach enables efficient parallel updates across all token dimensions and is agnostic to the training paradigm of the discrete diffusion prior, including masked and uniform-state diffusion. We evaluate our method on image restoration tasks across MNIST, CIFAR, and FFHQ, as well as spatial mapping, covering linear, nonlinear, and blind inverse problems. Across these settings, we improve over recent discrete diffusion posterior samplers and are competitive with strong continuous diffusion-based inverse solvers. Our results suggest that fully discrete, gradient-informed posterior samplers offer a scalable and general path toward solving inverse problems over discrete representations.

Discriminative models are trained with task-specific objectives such as classification, detection, or representation learning, yet their practical utility depends on generalizing beyond the training task to unseen samples from the same domain. This suggests that a discriminative model may retain an implicit trace of the data manifold beyond its task labels. We ask whether this trace can be exposed as a generative signal without task-specific guidance. We introduce the Discriminative Score Function (DSF), which extracts an image-space generative field from the loss-gradient geometry of a pretrained discriminative model. DSF provides a training-free update rule that moves images toward model-consistent visual regions without retraining or task-specific targets. We apply DSF to diverse pretrained discriminative models, including ResNet classifiers, DETR detectors, and DINOv2 ViT encoders. DSF generates unconditional images from noise and supports conditional guidance, editing, inpainting, and explanation through the same field. DSF-generated samples also provide effective zero-shot quantization calibration when real images or labels are unavailable, preserving feature geometry better than data-free calibration baselines. These results show that discriminative training leaves a generative functional signal for synthesis and calibration.

Disentangled representation learning aims to capture the underlying explanatory factors of observed data, enabling a principled understanding of the data-generating process. Recent advances in generative modeling have introduced new paradigms for learning such representations. However, existing diffusion-based methods encourage factor independence via inductive biases, yet frequently lack strong semantic alignment. In this work, we propose a flow-matching–based framework for disentangled representation learning, which casts disentanglement as learning factor-conditioned flows in a compact latent space. To enforce explicit semantic alignment, we introduce a non-overlap (orthogonality) regularizer that suppresses cross-factor interference and reduces information leakage between factors. Extensive experiments across multiple datasets demonstrate consistent improvements over representative baselines, yielding higher disentanglement scores as well as improved controllability and sample fidelity.

We characterise disentanglement for smooth generative pushforward models, such as in VAEs and GANs. For a generator/decoder $g:\\mathcal{Z}\\to\\mathcal{X}$ and factorised prior $p(z)=\\prod_i p_i(z_i)$, we define disentanglement as standard statistical independence expressed over the generated manifold: {the pushforward density} $p_\\mu = g_\\#p$ factorises into one–dimensional "seam" factors, each controlled by a distinct latent coordinate. We prove that $p_\\mu$ factorises according to the SVD of $g$'s Jacobian; that disentanglement equates to two conditions on $g$ (C1-C2); and that under such conditions the seam factors are identifiable, up to permutation and sign. In the special case of Gaussian ($\\beta$-)VAEs, we show via an identity how diagonal posteriors promote C1-C2, in expectation, explaining why disentanglement arises modulated by $\\beta$. Experiments illustrate this mechanism on Gaussian data, dSprites, and CelebA.

Multi-channel images encode heterogeneous channel semantics aligned over a shared spatial structure, which requires models to jointly capture globally shared structure and channel-specific variations. Recent multi-channel Vision Transformers attempt to address this challenge by augmenting the `[CLS]` token with memory tokens to organize multi-channel representations under varying channel compositions. However, we find that these context tokens collapse onto a few dominant channels under high inter-channel redundancy, producing redundant representations and a biased global summary that fails to reliably support composition-aware channel interpretation. We propose **MuCa-ViT (Multi-Channel Composition-Aware Vision Transformer)**, a unified framework that addresses these limitations through two coupled mechanisms. **Factorial Token Learning (FTL)** enforces token-level decorrelation among context tokens, encouraging the `[CLS]` token to capture globally shared semantics while memory tokens preserve complementary channel-specific cues. **Channel Composition-Aware Modulation (CAM)** uses the FTL-refined `[CLS]` as a semantic anchor to generate sample-wise modulation signals, which enable composition-aware interpretation of channels within a unified backbone rather than relying on static channel embeddings. Experiments on JUMP-CP, CHAMMI, and So2Sat show that MuCa-ViT outperforms prior multi-channel ViTs across all three benchmarks, and the magnitude of improvement aligns with the inter-channel redundancy of each dataset ($\rho \in [0.19, 0.77]$), validating the diagnostic analysis that motivates our design.

Causal representation learning for time series has developed strong identifiability results in discrete-time latent causal models, but identifiability in continuous-time latent stochastic differential equation (SDE) models remains largely open. We address this gap using environment-induced shifts in diffusion covariance. We study additive-noise latent SDEs observed through an unknown nonlinear diffeomorphism, with shared drift but environment-specific diffusion covariance. We show that two diagonal diffusion regimes with pairwise distinct coordinate-wise variance ratios identify the latent coordinates up to permutation and scaling, without any sparsity assumption on the drift. We first prove this result for linear Ornstein--Uhlenbeck systems and then extend it to general additive-noise latent SDEs. Under mild smoothness, the instantaneous drift-Jacobian causal graph is identifiable up to the same permutation. We propose a two-stage estimator for latent disentanglement and optional graph recovery; experiments on synthetic systems confirm the predicted identifiability boundary, and an application to Hardanger Bridge monitoring data illustrates the approach on real sensor trajectories.

Personalized image generation aims to synthesize text-driven images conditioned on reference images, while mainly casting the generation as image customization for foreground and style transfer for background. Previous arts of diffusion models suffers from the text misalignment with background for image customization and foreground for style transfer during the denoising process. Such facts, as we observed, rooted from the entanglement among hybrid frequency bands during the denoising process. In particular, the low-frequency band primarily captures color information, mid-frequency band encodes global structure, while high-frequency band corresponds to local texture. Besides, the early (high-noise) denoising stage mainly focuses on the low-and-mid frequency bands, while the late (low-noise) denoising stage progressively exhibits the high-frequency band, resulting in the dominance of low-and-mid over high frequency bands, especially when entangled within a single image reference, tends to reinforce each other at the early stage, hence the pitfall of overwhelming suppression of text prompt guidance. To address such salient limitation, in this paper, we study personalized generation based on dual references - customization and color and style reference - and propose a paradigm to disentangle these Dual image references within Frequency-aware Diffusion Models, dubbed Dual-FDM, to simultaneously tackle two crucial personalized image generation tasks: customization style transfer and color style transfer, by disentangling different frequency bands via mask strategy within frequency domain. For customization style transfer, we replace the mid-frequency band of the background in the style reference with that from the foreground of the customized reference. For color style transfer, we substitute the low-frequency band of the background in the style reference with that from both the foreground and background of the color reference. Both the substituted frequency bands are used as the key and value to reconstruct the query foreground and background of the denoised personalized image. We further compute the ratio of information entropy of the substituted frequency bands to adaptively modulate the denoising timesteps between the early and late stages. Extensive experiments validate the superiority of Dual-FDM over the state-of-the-art diffusion models for personalized image generation. Our code can be accessed from the supplementary material package.

Embodied AI systems can pass planning and still fail in execution: a delay, resource conflict, synchronization change, or candidate invalidation can turn a valid point schedule into broken commitments. We introduce FLECAR, which gives the executor a dispatchable coordination envelope instead of fixed start times. Offline, Flexible Envelope Synthesis builds executable windows, coupling constraints, and a dispatch graph; online, Controllability-Aware Repair updates that envelope after events and repairs only the affected region when dispatchability fails. On generated coordination instances with the same event streams and budgets, FLECAR improves post-event feasibility from 0.3880 to 0.8400, terminal success from 0.5800 to 0.8400, and normalized repair burden from 0.5357 to 0.2964 relative to legacy point-schedule repair. A point baseline with CAR-like repair closes part of the gap, but envelopes still retain higher post-event feasibility and lower repair burden, while pooled one-step survivability remains near parity. The result points to a simple execution principle: after disturbances, live windows, couplings, and alternatives serve the executor better than a point schedule rebuilt after the fact.

The Rashomon set paradigm seeks to uncover the full collection of near-optimal predictive models from a given function class. Access to this set allows users to better understand predictive multiplicity, interact with models, and select those that best satisfy domain-specific constraints. For sparse generalized additive models (GAMs), however, existing approaches that approximate the Rashomon set only apply to restricted subsets of features, overlooking the fact that equally accurate models can rely on many different feature subsets. We address this gap by representing the Rashomon set through a maximally diverse collection of models. We define a Metropolis-Hastings style approach for sampling maximally diverse GAMs from the Rashomon set. Experiments show that our methods can produce substantially more diverse solutions across all tested diversity metrics, providing users with actionable alternatives that maintain similar predictive performance.

Multi-path reasoning methods such as self-consistency (SC) sample $K$ reasoning paths and choose the most frequent answer. However, their gains quickly plateau as $K$ increases, and existing methods do not predict when this saturation will occur. We formalize multi-path LLM reasoning as a diversity combining problem from wireless communications: each path is a noisy channel observation, and pairwise path correlation limits the effective number of independent votes, creating a finite saturation ceiling. Generalized least squares (GLS) analysis shows that, under exchangeability, the optimal symmetric linear combiner of latent embeddings is uniform, supporting majority vote as the natural default in standard SC while leaving room for weighting or pruning under heterogeneous prompt-template branches. Across 5 models and 12 benchmarks, prompt-template diversity reduces path correlation in $55$ of $57$ valid cells, with the strongest effect on open-ended QA. We derive an Adaptive-K rule that uses a four-path pilot to select $K^*$, retaining $97$--$104\%$ of MV@$K{=}32$ accuracy across Math, QA, and NLU. Our code is hosted at \url{https://anonymous.4open.science/r/DiversityCombining-4905}.

Long-document understanding requires models to find and combine evidence across many pages, layouts, tables, figures, and charts. Existing retrieval-augmented systems usually select evidence from a static index before generation, while recent agentic systems add multi-turn tool use but often rely on frozen proprietary backbones whose behavior is set by prompts. We present DocAtlas, a system that treats long-document understanding as a mutable-state information-seeking process. We instantiate DocAtlas as a mutable document harness: an external environment that determines what document information is searched, read, stored, reviewed, and shown to the model at each step. Given a document and question, the harness exposes search, reading, note-taking, and review tools, maintains a hierarchical tree and note store, and updates both as the agent records evidence. DocAtlas combines self-improving retrieval, selective evidence access, and active working memory under a fixed context budget. The same harness supports inference-time use with large VLMs and end-to-end reinforcement learning for compact VLM agents. With GPT-5.4, DocAtlas reaches 71.4\% on MMLongBench-Doc, exceeding the human-expert reference of 65.8\%. A Qwen3.5-4B VLM trained with end-to-end RL in the DocAtlas environment reaches 63.7\%, compared with a 54.4\% direct-input baseline, showing that mutable document-harness design can improve compact document agents by a large margin.

A growing failure mode in agent evaluation and training is that models can achieve high evaluation scores by exploiting shortcuts instead of solving the intended task, producing \emph{deceptive} performance. This makes evaluation scores unreliable as measures of true task-solving ability. We propose CapCode, a framework for constructing coding datasets with randomized tests whose best achievable non-cheating performance is deliberately capped below one. This capped-performance design gives evaluation scores a clearer interpretation: scores substantially above the cap are implausible and therefore provide evidence of cheating. To prevent cheating, we propose CapReward, a reward design based on the CapCode principle to discourage optimization beyond the cap. Experiments across multiple datasets show that CapCode detects cheating while preserving performance ranking, and CapReward reduces cheating behavior, yielding models that better follow the intended task specification.

Evaluating whether Multimodal Large Language Models can produce trustworthy, verifiable reasoning over long, visually rich documents requires evaluation beyond end-to-end answer accuracy. We introduce DocScope, a benchmark that formulates long-document QA as a structured reasoning trajectory prediction problem: given a complete PDF document and a question, the model outputs evidence pages, supporting evidence regions, relevant factual statements, and a final answer. We design a four-stage evaluation protocol—Page Localization, Region Grounding, Fact Extraction, and Answer Verification—that audits each level of the trajectory independently through inter-stage decoupling, with all judges selected and calibrated via human alignment studies. DocScope comprises 1,124 questions derived from 273 documents, with all hierarchical evidence annotations completed by human annotators. We benchmark 6 proprietary models, 12 open-weight models, and several domain-specific systems. Our experiments reveal that answer accuracy cannot substitute for trajectory-level evaluation: even among correct answers, the highest observed rate of complete evidence chains is only 29\%. Across all models, region grounding remains the weakest trajectory stage. Furthermore, the primary difficulty stems from aggregating evidence dispersed across long distances and multiple document clusters, while an oracle study identifies faithful perception and fact extraction as the dominant capability bottleneck. Cross-architecture comparisons further suggest that activated parameter count matters more than total scale. The benchmark and code will be publicly released.

The main challenge of long-tailed deep clustering is that imbalanced class frequencies cause head classes to dominate the partitioning of the representation space, which in turn submerges or distorts the cluster structures of tail classes. Existing methods typically alleviate this issue through heuristic head-tail partitioning or fixed auxiliary clustering. However, such approaches rely on manually specified discrete granularities and therefore struggle to capture the continuous latent structure of real-world long-tailed data. To address this limitation, we propose DPLC, a Dirichlet Process guided Long-tailed Clustering method that adaptively models the latent sub-cluster structure from a nonparametric Bayesian perspective. Specifically, DPLC periodically extracts soft latent sub-clusters from self-supervised embeddings and leverages them as data-driven rebalancing signals to mitigate the clustering bias induced by head-class dominance. By requiring neither a predefined number of auxiliary clusters nor explicit binary head-tail partitioning, DPLC enables more flexible modeling of complex long-tailed distributions. Extensive experiments show that DPLC consistently outperforms existing methods across multiple long-tailed clustering benchmarks and a wide range of imbalance ratios.

Mapping reaction pathways and transition states (TS) is fundamental to chemistry but computationally expensive at scale. The minimum energy pathway (MEP) dictates reaction rates and mechanisms, yet recovering it via electronic-structure methods requires thousands of costly force evaluations. Recent generative models accelerate TS identification but require slow iterative inference and only predict isolated saddle-point snapshots, missing the continuous reaction trajectory. We introduce Drift-React, an $\mathrm{SE}(3)$-equivariant generative framework that predicts complete reaction pathways in a single forward pass from only reactant and product geometries. By shifting distribution evolution to training via a Sinkhorn-weighted drifting field, Drift-React eliminates both the iterative force evaluations of NEB-style methods and the sequential ODE/SDE integration of diffusion and flow matching models. Evaluated on the Transition1x and Halo8 datasets, our one-step model generates physically consistent MEPs that accurately capture energetic bottlenecks and enable arbitrary-resolution sampling along the reaction coordinate. For isolated TS prediction, Drift-React matches the sub-Ångström accuracy of state-of-the-art iterative models while delivering orders-of-magnitude acceleration, clearing a major computational bottleneck for large-scale reaction network exploration.

World models are increasingly used to reduce environment interaction when post-training vision-language-action (VLA) policies with reinforcement learning. Yet they are typically trained to predict pixels one step ahead, even though their role here is not merely to render visually faithful videos but to produce learning signals from imagined rollouts that improve the policy. We show that this mismatch is substantial: lower pixel MSE can produce worse policy gradients, and a world model trained for policy utility can yield better downstream policies despite 17\% higher pixel MSE on the next frame. We propose the Downstream-Return Imagination Learning Loop (DRILL), a bilevel framework that updates the world model to maximize the policy's return after an inner GRPO update on imagined rollouts. DRILL has two instantiations. DRILL-VWI is a closed-form surrogate, requiring no additional simulator rollouts, that reweights prediction errors by policy visitation, the group-relative GRPO advantage, and model confidence. DRILL-IMG is a full meta-gradient method that differentiates through the inner update using truncated Hessian-vector products. We show that DRILL-VWI recovers the local one-step component of the DRILL-IMG meta-gradient under score-compatibility and small-residual assumptions. Across five manipulation simulators and two VLA backbones, DRILL-IMG raises mean success rate by 13.1 points over WMPO and 7.9 points over RLVR-World, and complements RLVR pretraining rather than competing with it. More broadly, we believe video world models for VLA reinforcement learning can be productively trained by the policy gradients they produce, not only by pixel-level prediction.

Spatiotemporal intelligence in autonomous driving (AD) requires an agent to integrate multi-view observations into a coherent scene representation, maintain object continuity across viewpoints and time, and reason about spatial relations, interactions, and future dynamics. However, existing AD vision-language benchmarks largely focus on single-view, static, ego-centric, or single-source question answering, leaving it unclear whether current Vision-Language Models (VLMs) can truly construct and reason over dynamic driving scenes. We introduce DriveSpatial a benchmark of 15.6K human-verified QA pairs across 20 tasks from five large-scale AD datasets. DriveSpatial evaluates four abilities: Cognitive Scene Construction, Multi-view Relational Understanding, Temporal Reasoning, and Generalization. Unlike prior benchmarks, DriveSpatial is generated from a dynamic multi-relational scene graph that encodes object states, spatial relations, interactions, camera visibility, and temporal correspondences, enabling QA pairs that enforce genuine cross-view and spatiotemporal reasoning. Evaluating 15 representative VLMs reveals a substantial human–model gap: the strongest model trails humans by 28.4 points, with Cognitive Scene Construction emerging as the key bottleneck. Further diagnostics show that language-only prompting is insufficient, while explicit BEV grounding consistently improves performance. These results suggest that current VLMs lack the scene-construction ability needed for reliable spatiotemporal driving intelligence. DriveSpatial and its construction pipeline will be released to support future research.

While one-step text-to-image generators, a fast and deployment-friendly class of visual generative models, can produce samples with a single forward pass, existing preference-finetuning methods fail to simultaneously achieve efficient adaptation, reward-model agnosticism, and diverse generation. In this work, we propose Drifting Preference Optimization (DrPO), an online preference-finetuning method for one-step generators. Inspired by the recent work of Drifting Models, DrPO first uses the target reward model to construct an on-policy dipole reward model from positive and negative samples. This dipole model induces a preference drifting field in latent feature space, whose gradients are then used to optimize the generator without backpropagating through the target reward model. Empirically, we evaluate DrPO across multiple reward models and SD-Turbo and SDXL-Turbo one-step generators, including HPSv3 and GenEval as representative alignment benchmarks, and validate that DrPO can efficiently and effectively finetune one-step generative models. Preliminary experiments further suggest that this sample-based gradient synthesis can be extended to offline preference finetuning.

Time Series Anomaly Detection (TSAD) is critical for ensuring reliability in real-world systems such as industrial monitoring, healthcare, and online services. However, learning-based TSAD methods trained on a single normal-only TargetSet, namely the target dataset, often suffer from incomplete and biased knowledge of normality, causing domain-level false alarms when unseen but valid normal patterns appear at test time. A natural extension is to leverage DomainSet, composed of datasets from one domain, to enrich domain-level normality. Yet, our analysis shows that simply incorporating DomainSet could make domain-admissible normal variations and TargetSet-specific anomalies highly entangled in the learned representations, weakening anomaly separability. Motivated by human experts who establish normal patterns, calibrate admissible variations, and identify true anomalies, we propose dStructAD, a domain-level structured representation framework for time series anomaly detection, to address this challenge. Specifically, dStructAD first builds structured domain-level normality based on the Kolmogorov-Arnold network, and then calibrates admissible variations via two phases. Phase 1 learns shared domain-level semi-structural normality knowledge from DomainSet, while Phase 2 performs structure-preserving TargetSet calibration to absorb admissible variations without collapsing anomaly separability. Across six benchmarks, dStructAD delivers consistent improvements over SOTAs, with 3% average gain in overall performance. Code could be available at https://anonymous.4open.science/r/dStructAD-8F03.

Transfer learning seeks to improve efficiency in a target regression task by borrowing information from related external source dataset(s). Existing approaches achieve this by enforcing Euclidean proximity between the target and source regression parameters or, more recently, by encouraging their directional alignment through angle-based penalties. While such angle-based penalties mitigate the sensitivity of the transfer learning methods to the scale differences between source and target regression coefficients, they still rely on a point estimate of the source regression parameter and thus provide no direct mechanism to incorporate uncertainty in the source information. To that end, we propose a probabilistic framework for transfer learning that operates at the level of a \emph{scale-invariant directional distribution} of the source regression parameter estimate, rather than its point estimate, to enable robust, scale-invariant and uncertainty aware transfer under heterogeneous source information. Specifically, we project the sampling distribution of the source regression estimate onto the unit sphere, thereby extracting its scale-free directional distribution. Transfer from source to target is then induced through a novel \emph{angular cone prior} which shrinks the target regression parameter towards directions in which the source distribution assigns high probability, while allowing its magnitude to be learned exclusively from the target dataset. The proposed construction admits exact finite-sample characterizations of posterior angular concentration and posterior computation remains tractable via careful low-dimensional augmentation in the parameter space. Simulation studies and application on a benchmark hypertension prediction task based on the National Health and Nutrition Examination Survey (NHANES) datasets demonstrate that the proposed method yields marked gains over existing approaches in settings with substantial uncertainty in the source regression estimate and source--target scale mismatch.

Federated Domain Generalization (FDG) aims to learn a global model robust to heterogeneous domain shifts without sharing raw data, a critical challenge in multi-center medical imaging. Most existing methods implicitly assume static inter-domain discrepancies, overlooking the fact that client representations evolve continuously during local optimization. We identify this phenomenon as representation dynamics, which refers to the temporal evolution of latent features induced by training updates rather than shifts in underlying data distributions. Such dynamics often lead to unstable aggregation, particularly in heterogeneous medical imaging scenarios. To address this issue, we propose Dynamic Knowledge Tracked FDG (DKT-FDG), a framework that explicitly tracks these dynamics by leveraging flow matching to model continuous representation trajectories. By aligning representation evolution instead of static snapshots, DKT-FDG improves robustness to domain shifts. Comprehensive experiments on multiple multi-center medical benchmarks demonstrate consistent improvements over state-of-the-art FDG methods, highlighting the importance of modeling representation dynamics in federated learning.

Autoregressive Diffusion Transformers have emerged as a powerful paradigm for long video generation. However, they are often constrained by prohibitive computational overhead and substantial memory footprints. This inefficiency arises from dense computation mechanisms that allocate uniform processing resources across all spatiotemporal regions, failing to leverage the inherent temporal redundancy of video where motion is typically confined to sparse areas. In this paper, we propose **Dynamics-Aware Sparse Attention (DASA)**, a unified framework that shifts from dense computation to sparse, motion-driven generation to eliminate both storage and computational redundancies. Inspired by video codec principles, our approach explicitly decouples video content into dynamic and static regions. To mitigate memory bottlenecks, we selectively compress the historical KV cache by retaining only dynamic features within the generated chunk. Furthermore, to reduce computational overhead, we employ a region-level computation allocation strategy based on localized dynamics. Additionally, a dynamics-guided distillation method is introduced to enhance performance via a lightweight distillation training process. Experiments on VBench demonstrate that our approach accelerates autoregressive video generation by 2.16$\times$, while preserving high visual fidelity.

Vision–Language–Action (VLA) models offer a unified framework for robotic manipulation by integrating visual perception, language understanding, and control generation. However, existing VLA systems still struggle to generalize across diverse tasks, scenes, and camera viewpoints, and often produce coarse or unstable actions. We argue that these limitations are closely tied to the structural properties of actions in VLA settings, including the inherent multi-peaked nature of action distributions, the token-based symbolic reasoning of pretrained VLM/VLA backbones, and the effective finite resolution imposed by real-world robotic control. Motivated by these properties, we introduce E0, a tweedie discrete diffusion framework that formulates action generation as iterative denoising over quantized action tokens. By operating in a discrete action space with a principled diffusion process, E0 naturally aligns with token-based reasoning, supports fine-grained yet executable action control, and avoids the distributional mismatch of masking-based discrete diffusion. Experiments on LIBERO, VLABench, ManiSkill, and a real-world Franka arm demonstrate that E0 achieves state-of-the-art performance across diverse environments, outperforming strong baselines by 8.3\% on average.

Pretrained video diffusion models provide powerful spatiotemporal generative priors, making them a natural foundation for robotic world models. While recent world-action models jointly optimize future videos and actions, they predominantly treat video generation as an auxiliary representation for policy learning. Consequently, they insufficiently explore the inverse problem: leveraging action signals to guide video synthesis, thereby often failing to preserve precise robot spatial geometry and fine-grained robot-object interaction dynamics in the generated rollouts. To bridge this gap, we present EA-WM, an Event-Aware Generative World Model that effectively closes the loop between kinematic control and visual perception. Rather than injecting joint or end-effector actions as abstract, low-dimensional tokens, EA-WM projects actions and kinematic states directly into the target camera view as Structured Kinematic-to-Visual Action Fields. To fully exploit this geometrically grounded representation, we introduce event-aware bidirectional fusion blocks that modulate cross-branch attention, capturing object state changes and interaction dynamics. Evaluated on the comprehensive WorldArena benchmark, EA-WM achieves state-of-the-art performance, outperforming existing baselines by a significant margin.

In time-series forecasting, rollout errors are not merely one-step fitting discrepancies: once fed back into the evolving state, they can perturb future predictions. We formalize this effect through local rollout consistency, which compares the next prediction produced under teacher forcing with that produced under free running. In principle, this consistency can be enforced directly by tracking the residual-induced perturbation of the next state. In practice, however, the residual-to-state alignment operator is often difficult to design when the input state and output block have different dimensions, heterogeneous components, or architecture-dependent layouts. We therefore use a first-order local surrogate that penalizes the Jacobian-vector product of the predictor along the actual feedback perturbation, suppressing precisely the directions through which self-generated errors remain visible to future predictions. We further give the same loss a geometric interpretation near a predictive manifold: by controlling residual-induced directions that probe the normal bundle, the objective yields theoretical conditions under which free-running trajectories remain inside a stability tube around the manifold. Experiments on time series forecasting tasks support this view, showing that the proposed regularization improves predictive accuracy. More broadly, our results suggest a training principle for time-series forecasters and sequential predictors: a model should not only make small one-step errors, but also learn not to let its own errors echo into the future.

Long-form video understanding remains challenging for Video Large Language Models (VideoLLMs), as the dense frame sampling introduces massive visual tokens while sparse sampling risks missing critical temporal evidence and leading to LLM hallucination. Existing training-free token reduction methods either treat videos equally as static images or rely on segment-level merging heuristics, which weaken fine-grained spatiotemporal modeling and introduce additional overhead. In this paper, we propose EchoPrune, a lightweight and training-free token pruning method that improves temporal resolution under a fixed LLM-side visual token budget. Our core idea is to interpret redundant video tokens as temporal echoes: if a token is well reconstructed from the previous frame, it is merely a temporally redundant echo; otherwise, it may capture new events, motion, or query-relevant visual evidence. Based on this insight, EchoPrune scores visual tokens by (i) query-guided crossmodal relevance and (ii) temporal reconstruction error, measured by correspondence matching and echo matching across consecutive frames. The selected tokens preserve task-relevant cues and temporal novelty while suppressing predictable redundancy, allowing VideoLLMs to observe more frames without increasing the decoding budget. Extensive experiments on LLaVA-OV, Qwen2.5VL, and Qwen3VL across six video understanding benchmarks show that EchoPrune enables VideoLLMs to process up to $\mathbf{20\times}$ frames under the same token budget, yielding improved performance ($\mathbf{+8.6\\%}$) $\%$and inference speedup ($\mathbf{5.6\times}$ for prefilling) on Qwen2.5VL-7B.

Deep learning has become the dominant paradigm for vision-based spacecraft pose estimation, where models are predominantly trained on synthetic data due to the prohibitive cost of acquiring real in-orbit imagery. When deployed, these models suffer significant performance degradation due to the sim-to-real domain gap, which existing benchmarks address by pairing synthetic training data with Hardware-in-the-Loop (HIL) real imagery. Yet a critical factor remains consistently overlooked: in real orbital scenarios, illumination varies with the spacecraft's position along its orbit, inducing severe appearance shifts that existing benchmarks neither control nor study. We introduce ECLIPSE, the first benchmark designed to isolate illumination as an independent experimental variable. ECLIPSE pairs a large-scale photorealistic synthetic training set with a fully labeled HIL evaluation set of 14 approach trajectories, each replayed identically under 4 controlled lighting conditions, enabling rigorous evaluation of illumination-robust and domain-adaptation methods. Through this controlled design, our experiments confirm that lighting is the dominant driver of the sim-to-real performance gap in spacecraft pose estimation.

Web-enabled LLM agents are changing how online information influences search outcomes. Existing Generative Engine Optimization (GEO) studies mainly focus on individual webpages. However, agentic web search is not a single-document setting: an agent may issue queries, crawl pages, follow links, reformulate searches, and synthesize evidence across multiple browsing steps. Influence therefore depends not only on page content, but also on how pages are organized, connected, and encountered along the agent's browsing trajectory. We study this shift through Ecosystem Generative Engine Optimization (EcoGEO), which treats GEO as an environment-level influence problem for web-enabled LLM agents. To instantiate this perspective, we propose TRACE, a Trajectory-Aware Coordinated Evidence Ecosystem. Given a recommendation query and a fictional target product, our method builds a controlled evidence environment that coordinates an agent-facing navigation entry page with heterogeneous support pages. These pages use shared terminology, internal links, and consistent product attributes to introduce, verify, and reinforce the target product. We evaluate our method on OPR-Bench, a benchmark for open-ended product recommendation. Experiments show that it consistently outperforms page-level GEO baselines in final target recommendation. Trajectory-level metrics further show increased initial target-result crawls, target-specific follow-up searches, and internal-link crawls, suggesting that the gains come from shaping the agent's evidence-acquisition process rather than merely adding more target-related content. Overall, our findings support an ecosystem research paradigm for GEO, where web-enabled LLM agents are studied in relation to the broader evidence environments that guide search, browsing, and answer synthesis.

Long-horizon planning is widely recognized as a core capability of autonomous LLM-based agents; however, current evaluation frameworks suffer from being largely episodic, domain-specific, or insufficiently grounded in persistent economic dynamics. We introduce EcoGym, a generalizable benchmark for continuous plan-and-execute decision making in interactive economies. EcoGym comprises three diverse environments: Vending (adapted from the closed-source Vending-Bench, with full open-source release), Freelance (new), and Operation (new), implemented in a unified decision-making process with standardized interfaces, and budgeted actions over an effectively unbounded horizon (1000+ steps if 365 day-loops for evaluation). The evaluation of EcoGym is based on business-relevant outcomes (e.g., net worth, income, and DAU), targeting long-term strategic coherence and robustness under partial observability and stochasticity. Experiments across eleven leading LLMs expose a systematic tension: no single model dominates across all three scenarios. Critically, we find that models exhibit significant suboptimality in either high-level strategies or efficient actions executions. EcoGym is released as an open, extensible testbed for transparent long-horizon agent evaluation and for studying controllability–utility trade-offs in economic settings.

Symbolic regression (SR) aims to discover compact, interpretable mathematical expressions from data. This goal inherently requires iterative refinement rather than one-shot construction. Autoregressive methods lack the ability to perform targeted corrections, so whenever structural flaws appear, the remainder of the sequence must be discarded and regenerated. Meanwhile, search-based methods rely on costly and indirect replacement through enumeration. We propose EditFlowSR, which reframes SR as iterative editing of expression trees through insertions, deletions, and substitutions. To reconcile fitting accuracy with algebraic simplicity, we introduce dual trajectory supervision. One trajectory teaches the model to construct expressions from data, while the other teaches it to algebraically simplify them, and both are unified under the shared editing framework. At inference time, EditFlowSR starts from a randomly initialized expression and progressively refines it through iterative editing, simultaneously constructing valid structure and removing redundancy. Across the standard SRBench, EditFlowSR achieves first-rank Pareto dominance in predictive accuracy, expression tree size, and time complexity. These results establish edit-based generation as a principled and effective paradigm for end-to-end symbolic regression. Code is available at \url{https://anonymous.4open.science/r/EditFlowSR_NIPS2026}.

Multimodal question answering (MMQA) is based on integrating heterogeneous data sources, selectively leveraging relevant modalities, and ignoring distractors. Existing approaches based on Multimodal Large Language Models (MLLMs) improve performance by splitting the task into intermediate steps, but do not quantify how each modality contributes to reaching the final answer. We argue that quantifying the effect that each modality has on the answering process improves process traceability and final performance. Thus, we propose Entropy-Driven Multimodal Answering (EDMA), which uses logical entropy to formalise multimodal QA as an iterative process of entropy reduction and to quantify the contribution of each modality in distinguishing correct from incorrect candidate answers. This enables the construction of quantifiable answering trajectories, where each step is associated with a measurable entropy reduction. EDMA introduces (i) modality-conditioned partitioning to estimate unimodal entropy reduction, (ii) entropy-driven multimodal fusion to capture complementary information across modalities, and (iii) entropy-based selection of the answering trajectory that minimises the most logical entropy. Experiments on multimodal QA benchmarks show that EDMA consistently outperforms prompting-based baselines and state-of-the-art methods. Averaged across datasets, EDMA achieves gains of $12.8$ and $19.8$ F1 points over the strongest prompting baseline and state-of-the-art system, respectively. Moreover, we show that higher entropy correlates with more false positives and, through controlled interventions on the choice of the answering trajectory, provide evidence that lower-entropy partitions causally reduce false positives, validating entropy level as a reliable proxy for performance. We share our code at https://anonymous.4open.science/r/EDMA/README.md.

We seek to learn discrete skill abstractions to facilitate long-horizon, goal-conditioned reinforcement-learning (RL) tasks. Recent approaches have shown success by quantizing actions via clustering or reconstruction. However, these objectives lead to skills that are not directly related to what the current state affords or what the goal requires. Instead, we argue that goal-directed skills should be driven by how they affect the state of the agent. To realize this, we introduce a novel quantization method for action chunks that optimizes next-state predictions conditioned on the current state and action. This objective encourages skills to be distinguished by the future states they lead to, resulting in a vocabulary that spans diverse outcomes reachable from a given state. Our framework can easily be paired with common offline RL methods by replacing their continuous actor head with a discrete one that selects among the learned skills. We can execute these skills in the environment by converting them back to continuous actions using a trained low-level controller. Experiments on benchmark navigation tasks largely show substantial and consistent improvements over both the continuous-actor baselines and reconstruction-based discretizations.

While animals and people tend to learn a task more quickly and reliably when they first train on simpler versions of it, curriculum learning's effectiveness in artificial settings varies, and appears substantially greater in reinforcement learning than in supervised learning. In this work, we construct and analyze a minimal model of policy learning to better understand why. We consider an open-loop, episodic target interception task whose parameters---especially how close the agent must be to the target to receive an informative reward signal---can be chosen so that the agent either receives informative rewards frequently ('dense' rewards), or rarely ('sparse' rewards). We mathematically and numerically analyze the dynamics of REINFORCE-based policy learning in both cases. In the dense reward case, curricula have only marginal benefits; in the sparse case, curricula can be required for learning to occur at all. We find that this is especially true when the agent has the conflicting goals of minimizing effort and accumulating high reward, since this can produce distinct reward landscape optima which compete with one another. Our characterization of useful curricula in this setting indicates that they address two issues: they (i) make rewards less sparse to speed up learning, and (ii) shape the reward landscape to steer agents away from bad local optima.

Modern data workflows are inherently adaptive, repeatedly querying the same dataset to refine and validate sequential decisions, but such adaptivity can lead to overfitting and invalid statistical inference. Adaptive Data Analysis (ADA) mechanisms address this challenge; however, there is a fundamental tension between computational efficiency and sample complexity. For $T$ rounds of adaptive analysis, computationally efficient algorithms typically incur suboptimal $O(\sqrt{T})$ sample complexity, whereas statistically optimal $O(\log T)$ algorithms are computationally intractable under standard cryptographic assumptions. In this work, we shed light on this trade-off by identifying a natural class of data distributions under which both computational efficiency and optimal sample complexity are achievable. We propose a computationally efficient ADA mechanism that attains optimal $O(\log T)$ sample complexity when the data distribution is dense with respect to a known prior. This setting includes, in particular, feature-label data distributions arising in distribution-specific learning. As a consequence, our mechanism also yields a sample-efficient (i.e., $O(\log T)$ samples) statistical query oracle in the distribution-specific setting. Moreover, although our algorithm is not based on differential privacy, it satisfies a relaxed privacy notion known as Predicate Singling Out (PSO) security (Cohen and Nissim 2020). Our results thus reveal an inherent connection between adaptive data analysis and privacy beyond differential privacy.

LLM agents can optimize GPU kernels, but each candidate requires generation, compilation, correctness testing, and profiling. Our objective is to reduce the number of expensive attempts needed to find a correct, fast kernel. We study two complementary ways to improve this attempt efficiency by changing the representation that the agent writes in and adding a first-principles performance signal that estimates remaining headroom. We introduce $\mu$CUTLASS, a compact in-context learnable DSL that exposes high-impact CUTLASS optimization choices while hiding template plumbing, and pair it with Speed-of-Light (SOL) guidance for search steering, budget allocation, and integrity checking. On 59 KernelBench problems on H100, switching GPT-5-mini from low-level code generation to $\mu$CUTLASS changes a $0.40\times$ geomean regression versus PyTorch into a $1.27\times$ speedup. Adding SOL-guided steering, it reaches $1.56\times$. Across model tiers, this combination lets weaker models match or exceed stronger raw-code baselines, and SOL-guided scheduling saves 19--43\% of tokens while retaining at least 95\% geomean speedup. We also show that integrity filtering is essential because, without SOL-assisted checks, reported speedups can be inflated by up to $1.9\times$ by benchmark-gaming or PyTorch-only solutions.

Data mixing is a consequential problem throughout language model training. In pretraining, data composition is a key determinant of model quality; in continual learning and adaptation, it governs what is retained and acquired. Yet existing data mixing methods address only one phase of this lifecycle at a time: some require smaller proxy models tied to a single training phase, others assume a fixed domain set, and continual learning lacks principled guidance altogether. We argue that data mixing is fundamentally an online decision making problem---one that recurs throughout training and demands a single, unified solution. We introduce OP-Mix (On-Policy Mix), a data mixing algorithm that operates across the entire language model training lifecycle. Our main insight is that candidate data mixtures can be cheaply simulated by interpolating between low-rank adapters trained directly on the current model, eliminating separate proxy models and ensuring the search is always grounded in the model's actual learning dynamics. Across pretraining, continual midtraining, and continual instruction tuning, OP-Mix consistently finds near-optimal mixtures while using a fraction of the compute of the baselines. In pretraining, OP-Mix improves upon training without mixing by 6.3% in average perplexity. For continual learning, OP-Mix matches the performance of both retraining and on-policy distillation while using 66% and 95% less overall compute, respectively. OP-Mix suggests a different view of language model training: not a sequence of distinct phases, but a single continuous process of learning from data.

Brain-to-speech decoding aims to restore communication by translating neural activity into speech-related representations. From a modeling perspective, this task can be formulated as a weakly supervised neural sequence decoding problem, where phoneme sequences must be recovered without frame-level annotations. Existing speech decoders commonly rely on recurrent sequence models to integrate temporal context, but their dense sequential computation can limit their suitability for resource-constrained portable or implantable neuroprosthetic systems. In this study, we propose a residual spiking neural network (SNN) for phoneme-level decoding of intracortical speech signals under constrained model capacity and weak sequence-level supervision. Our approach incorporates biologically inspired fixed connection-specific delays to integrate temporal context, providing each postsynaptic unit with access to recent presynaptic spike activity without recurrent state or dense temporal convolutions. Because each directed connection is assigned a single fixed delay, the temporal field is expanded without adding trainable delay parameters. Across 500k, 2M, and 5M core-parameter budgets, fixed delays consistently improve phoneme error rate (PER), with gains saturating once the delay range covers the task-relevant temporal horizon. Boundary-aligned CTC analysis further suggests that delayed temporal context reduces local uncertainty around phoneme transitions, indicating that fixed delays support weakly supervised phoneme alignment rather than simply increasing model capacity. Compared with recurrent, convolutional, and feedforward baselines under matched core-parameter budgets, the proposed SNN achieves the best PER at 500k and 2M parameters and the second-best PER at 5M parameters, while reducing estimated operation-level energy by approximately 2-4$\times$ relative to the strongest non-spiking baselines. These results suggest that fixed-delay SNNs provide an efficient temporal modeling mechanism for brain-to-speech decoding, potentially enabling lightweight decoding for on-device speech neuroprostheses.

Generalized planning aims to learn policies that generalize across large collections of instances within a classical planning domain. Recent approaches using Graph Neural Networks (GNNs) have shown the ability to learn nearly perfect policies for several domains. This work improves on the recently published idea of Iterated Width (IW) policies. Therein, the policy broadens its successor scope through an IW-lookahead search that offers to "jump" over multiple transitions, simplifying the problem structure. Yet, each transition is evaluated individually leading to unscalable compute and expressivity limitations. Furthermore, while an IW(1) search is attractive due to scaling linearly with the number of atoms in a problem, it still becomes inefficient once thousands of objects are considered like in the International Planning Competition (IPC) 2023 benchmark. In this work, we address both limitations. Firstly, we introduce a vastly more efficient holistic encoding of the entire search tree. It jointly represents IW(1)-reachable states only by their relational differences to the current state, which enables Relational GNNs (R-GNNs) to score all transitions in a single forward pass. Secondly, we define Abstracted IW(1) to improve scaling through relational abstraction during novelty checks. Rather than testing fully instantiated atoms, it abstracts each atom by replacing all but one of its arguments with their types. The original atom is then novel if any of its abstracted forms is. This structural compression shifts the scaling of the novelty search to be linear in the number of objects, rather than atoms, still capturing meaningful subgoal structure. Our contributions are evaluated on the hyperscaling IPC 2023 benchmark and across a broad range of domains, including those that require features beyond the $C_2$ logic fragment. The results show that our policies achieve a new state-of-the-art performance, significantly surpassing prior work, including classical planners like LAMA.

Pre-training of Large Language Models is often prohibitively expensive and inefficient at scale, requiring complex and invasive modifications in order to achieve high data throughput. In this work, we present Token-Superposition Training (TST), a simple drop-in method that significantly improves the data throughput per FLOPs during pre-training without modifying the parallelism, optimizer, tokenizer, data, or model architecture. TST is done in two phases: (i) A highly efficient superposition phase where we combine many contiguous tokens into one bag and train using a multi-hot cross-entropy (MCE) objective, and (ii) a recovery phase where we revert back to standard training. We extensively evaluate TST on the scale of 270M and 600M parameters and validate on 3B and a 10B A1B mixture of experts model, demonstrating that it is highly robust in different settings. Ultimately, TST consistently outperforms baseline loss and downstream evaluations, and under equal-loss settings, TST yields up to a 2.5x reduction in total pre-training time at the 10B A1B scale.

Fast and accurate retrosynthesis prediction is desired for downstream drug discovery and synthetic planning tasks. Currently, training and sampling state-of-the-art discrete diffusion or flow-based models for retrosynthesis requires significant computational resources. In this work, we propose FlashRetro, an Integral Flow Matching (IFM) framework with Latent Inversion for efficient retrosynthesis prediction. FlashRetro learns finite-interval latent transport rather than local instantaneous velocity, directly matching the transport quantity required for one-step latent inversion and avoiding test-time numerical integration. For inference, FlashRetro uses a principled 1-NFE Latent Inversion rule that maps a noise latent conditioned on the product latent to the reactant latent with a single learned integral-transport update. Experimental results on standard benchmarks (USPTO-50K) show that FlashRetro achieves superior top@1 accuracy, outperforming prior discrete flow-matching methods by 13.5\% and diffusion-based baselines by 14.2\%, while its average inference time is only 5.1 ms per reactant---38$\times$ faster than multi-step diffusion-based models and nearly 22$\times$ faster than recent discrete flow models. These results show that FlashRetro provides an effective path toward highly efficient 1-NFE retrosynthesis prediction with flow-based models.

Spiking neural networks (SNNs) are a promising paradigm for efficient neuromorphic computing, yet their training remains challenging due to the non-differentiability nature of spike firing. Backpropagation through time (BPTT) with surrogate gradients (SG) has become a mainstream approach due to its superior performance. However, this method incurs high computational and memory demands during training, while its update rule depends on specific neuronal dynamics. Furthermore, when targeting neuromorphic hardware implementation, the temporal error-propagation process and the full membrane-potential access required by SG are difficult to realize; these can introduce additional readout overhead, measurement noise, or perturbations to the system dynamics. To address these issues, we propose the input-driven derivative-free training (IDDFT) method. Rather than relying on membrane-potential-based surrogate derivatives, IDDFT constructs derivative-free error signals by applying a relaxed nonlinearity to the inputs, thereby avoiding the temporal error-propagation process and reducing the dependence of the update rule on specific neuronal dynamics. By eliminating access to internal membrane-potential states, our method reduces computational and memory costs, enhances compatibility with neuromorphic hardware, and enables training of black-box SNN models. We validate the IDDFT method through theoretical analysis and systematic experiments, demonstrating its effectiveness as an input-driven, derivative-free mechanism for constructing update signals. When integrated with a biologically inspired training strategy, IDDFT maintains comparable performance while substantially enhancing robustness to the choice of relaxed nonlinearities. Based on the resulting bio-inspired combined framework, we conduct evaluations on CIFAR-10, CIFAR-100, CIFAR10-DVS, and Tiny-ImageNet. Experimental results demonstrate that our method achieves performance comparable to BPTT and state-of-the-art methods, while exhibiting remarkable robustness under adversarial attacks.

Speculative decoding is an effective technique for accelerating autoregressive models in memory-bound regimes. Prior work has explored tree-based drafting to speculate multiple candidates at each step, improving verification acceptance rates. However, existing systems fail to fully realize the potential of tree-based drafting, leaving significant performance overhead. In this paper, we present FastDraft, a plug-and-play module that improves the efficiency of tree-based drafting in long-context speculative decoding. FastDraft exploits both the shared-prefix I/O pattern across draft branches and the constrained tree structure of speculative decoding, enabling specialized kernels that reduce draft-phase overhead while preserving compatibility with existing inference frameworks. FastDraft requires no modifications to the target model, draft model, or sampling logic. We implement FastDraft within SGLang and compare it against SGLang’s vanilla tree-based drafting backend, demonstrating substantially faster draft execution: up to 2.66× speedup in the draft phase and 1.77× higher throughput over SGLang under long-context settings.

As AI agents increasingly operate in open, real-world environments, they require a deep synergy of multimodal perception, tool invocation with multi-hop reasoning, and dynamic interaction with users. However, existing benchmarks fail to jointly evaluate these capabilities due to challenges in designing strictly coupled multi-capability tasks, simulating natural and task-constrained user feedback, and ensuring objective evaluation of dynamic interaction. To bridge this gap, we introduce EgoBench, the first interactive multimodal benchmark for tool-using agents. EgoBench comprises 1,045 egocentric-video-grounded tasks covering four daily scenarios, along with a user–agent–tool interactive environment for evaluation. We implement a three-stage synergistic pipeline through which each task is designed to enforce the joint application of visual perception and tool-augmented multi-hop reasoning. We additionally develop a multi-agent simulated user within EgoBench to evaluate agents' interaction capabilities, which generates high-fidelity, task-aligned responses to agents. Furthermore, we establish a deterministic joint validation framework that guarantees objective assessment through process-based and result-based equivalence. Benchmarking eight SOTA video-MLLM agents on EgoBench reveals a severe performance ceiling: the best model achieves only 30.62% accuracy in the best-performing scenario, averaging 19.43% across all four scenarios. Finally, we conduct a multi-dimensional error analysis to disentangle failure modes, exposing capability bottlenecks for advancing future AI agents.

The complex interactive dependencies between human behavior and 3D scenes make accurately predicting human intent a major challenge for human motion prediction (HMP). While recent studies use human gaze to infer intent, gaze signals suffer from ambiguity, drift, and a reliance on eye-tracking hardware. Research in cognitive science indicates that egocentric views inherently encode potential interactions, while head motion serves as a crucial precursor in motion planning. Motivated by this, we propose EgoHMP, a novel framework that uses egocentric views and head motions as robust carriers of interaction intent. By translating this intent into an envisioning of future interactions, EgoHMP achieves precise HMP in 3D scenes. EgoHMP first employs a vision-motion alignment encoder to align the visual and motion feature spaces. Based on these aligned features, an egocentric intent predictor with a modality-aware modulation mechanism adaptively suppresses invalid interaction cues to predict an accurate interaction probability map. Furthermore, to enhance the stability and realism of the generated motions, we propose a contact-aware motion decoder. Specifically, inspired by the human cognitive process of planning trajectories prior to action execution, we first predict a global trajectory to serve as a self-prompt, effectively mitigating the optimization conflict in diffusion models. Subsequently, we introduce a multi-stage denoiser, which relieves motion artifacts by incorporating human contact modeling through a multi-stage optimization strategy. Finally, a GCN-based motion decoder is employed to synthesize physically plausible and semantically consistent motions. Extensive experiments demonstrate that EgoHMP achieves state-of-the-art performance.

Discharge summaries are crucial clinical documents for patient care, containing clinical context of a patient's overall admission, and are routinely reviewed by medical experts during patient readmission, ongoing care, and diagnostic decision-making. In practice, medical experts often must synthesize information across multiple lengthy discharge summaries by iteratively obtaining information from the documents, while also verifying the evidence supporting each answer. Although large language models (LLMs) are increasingly explored for clinical document question answering, existing benchmarks do not sufficiently capture this clinical setting: they either evaluate exam-style medical knowledge or focus on single-turn question answering with limited evidence-grounding evaluation. We introduce EHRNote-ChatQA, the first benchmark for evidence-grounded multi-turn clinical question answering over patients' multiple discharge summaries. Built from de-identified MIMIC-IV discharge summaries, EHRNote-ChatQA contains 967 patient-level multi-turn samples spanning one to five notes and 16,072 medical-expert-verified QA pairs (8,036 content questions, each paired with an evidence-grounding question) across eight clinical categories. We construct the benchmark through an expert-informed pipeline that combines a comprehensive discharge-summary structuring schema, expert-curated multi-turn QA templates, and LLM-based generation. Every single QA sample is then reviewed and revised by 11 medical experts over three weeks. Benchmarking 22 LLMs (both open and closed-sourced) reveals several findings; evidence grounding is consistently harder than content answering, multi-turn errors compound across turns, and strong performance on existing single-turn clinical QA benchmarks does not reliably transfer to this setting. These results establish EHRNote-ChatQA as a rigorous and practical benchmark for evaluating clinical QA systems.

Diffusion and flow-based models have become the de facto approaches for generating continuous data, e.g., in domains such as images and videos. Their success has attracted growing interest in applying them to language modeling. Unlike their image-domain counterparts, today’s leading diffusion language models (DLMs) primarily operate over discrete tokens. In this paper, we show that continuous DLMs can be made effective with minimal adaptation to the discrete domain. We propose Embedded Language Flows (ELF), a class of diffusion models in continuous embedding space based on continuous-time Flow Matching. Unlike existing DLMs, ELF predominantly stays within the continuous embedding space until the final time step, where it maps to discrete tokens using a shared-weight network. This formulation makes it straightforward to adapt established techniques from image-domain diffusion models, e.g., classifier-free guidance (CFG). Experiments show that ELF substantially outperforms leading discrete and continuous DLMs, achieving better generation quality with fewer sampling steps. These results suggest that ELF offers a promising path toward effective continuous DLMs.

Humanoid robots, as general-purpose physical agents, must integrate both intelligent control and adaptive morphology to operate effectively in diverse real-world environments. While recent research has focused primarily on optimizing control policies for fixed robot structures, this position paper argues for “evolving both control strategies and humanoid robots' physical structure under a co-design mechanism”. Inspired by biological evolution, this approach enables robots to iteratively adapt both their form and behavior to optimize performance within task-specific and resource-constrained contexts. Despite its promise, co-design in humanoid robotics remains a relatively underexplored domain, raising fundamental questions about its feasibility and necessity in achieving true embodied intelligence. To address these challenges, we propose practical co-design methodologies grounded in strategic exploration, Sim2Real transfer, and meta-policy learning. We further argue for the essential role of co-design by analyzing it from methodological, application-driven, and community-oriented perspectives. Striving to guide and inspire future studies, we present open research questions, spanning from short-term innovations to long-term goals. This work positions co-design as a cornerstone for developing the next generation of intelligent and adaptable humanoid agents.

Comparing object orientations and positions across different instances requires their poses to be expressed in a shared canonical frame. Establishing such frames has traditionally required manual annotation, creating a scaling bottleneck that limits category and instance diversity. We show that a shared canonical frame can instead emerge from self-supervised training on object-centric videos captured in the wild, using only noisy camera poses from Structure-from-Motion. Our key idea is to route all training sequences through a shared geometric bottleneck: a coarse canonical mesh that carries no category-specific detail. By learning dense correspondences from image pixels to this mesh, and estimating per-sequence alignments from noisy SfM geometry, a common canonical frame emerges from multi-view consistency and the semantic priors of the feature extractor, without any canonical pose labels or category conditioning. Trained in a self-supervised manner on $160{,}000$ in-the-wild object videos, our method achieves competitive accuracy on category-level pose estimation benchmarks compared to methods that rely on canonical pose supervision.

Foundation models trained on large unlabeled corpora develop emergent capabilities — tasks that go beyond the training objective. We extend this paradigm to multimodal biology with Bonbon, a foundation model for molecular interactions trained on approximately 3.2 trillion tokens of paired protein-ligand sequences. From a single self-supervised objective, Bonbon exhibits four zero-shot emergent capabilities at hierarchical levels of resolution — functional (mechanism of action), residue (binding site localization, including orthosteric/allosteric discrimination), bond (covalent versus non-covalent engagement), and atom (active moiety identification at single-bond resolution). All four capabilities generalize to out-of-distribution proteins. Bonbon was trained on protein-ligand sequences alone, without any task-specific supervision and without structural coordinates. We conducted controlled scaling experiments across four model scales from 42M to 1.6B parameters; every capability evaluated shows a sharp jump at the largest scale on continuous metrics, even as pretraining loss decreases smoothly. This demonstrates that loss-based scaling laws alone are insufficient to predict when zero-shot emergent capabilities become available. The capabilities reported here emerge after training at a 1:2,000 parameter-to-token ratio to accommodate the sparsity of paired biological interaction data. The same frozen representations achieve state-of-the-art binding and affinity predictions at five orders of magnitude greater speed than structure-based methods. In prospective wet-lab validation, Bonbon achieves a 70% hit rate on de novo compounds, discovering novel chemotypes not previously documented as inhibitors for those targets. All of the tested compounds had less than 60% structural similarity to molecules in the training corpus.

We introduce a three-wave, in-the-wild multimodal dataset for affect sensing that integrates smartphone sensing, wearable sensing, and dense experience-sampling-method (ESM) labels collected annually from 2020 to 2022. The dataset supports moment-level affect modeling through a shared dimensional label core across all waves, with additional affective descriptors available in the third wave (D-3). We describe the resource in terms of study design, temporal density of in-situ labels, and sensing and label coverage across waves. To support evaluation within this resource, we define an initial three-setting benchmark spanning temporal prediction from within-user history, within-wave cross-user generalization, and cross-wave generalization in which each wave is treated as a separate dataset. Our benchmark results show that the strongest method family depends on the evaluation setting: supervised baselines perform best in the temporal setting, unsupervised domain adaptation is strongest overall in the within-wave cross-user setting, and domain generalization shows the strongest overall cross-wave performance, although its margin over strong baselines is modest. These findings indicate that robust mobile affective computing is constrained not only by label availability but also by substantial participant-level variability and realistic cross-wave differences inherent in longitudinal in-situ deployments.

We address the problem of learning generative priors over high-dimensional latent variables from indirect, noisy observations, with a focus on continuous heterogeneity in cryo-electron microscopy (cryo-EM). In this setting, 3D atomic structures $x \in \mathbb{R}^{3N_a}$ are never directly observed and must be inferred from 2D projection images $y$ generated by a complex, non-linear forward operator. We propose Empirical Bayes Flow Matching (EB-FM), a method that learns a flow matching prior $p_\theta(x)$ directly from observations by embedding it within a Monte Carlo stochastic approximation Expectation-Maximization framework. EB-FM alternates between an E-step that performs approximate posterior sampling from $p_\theta(x \mid y)$ via guided flow-based generation while maintaining persistent latent estimates through stochastic approximation, and an M-step that updates the flow model using these stabilized latent variables as training targets. This approach eliminates the need for pretraining on clean data and naturally accommodates non-linear forward models. We validate our method on controlled inverse problems using two-moons and MNIST data, and demonstrate its effectiveness on simulated cryo-EM data with known poses, showing that EB-FM can recover continuous distributions of protein conformations directly in atomic-coordinate space.

When large language models (LLMs) are used for creative tasks where the evaluation is largely subjective, two crucial activities are ideation (in which a list of candidate ideas is produced) and pruning (in which the initial candidates are prioritized through a process of comparison). In this framework, the subjective results are influenced by inherent biases in the LLM's generation and evaluation, and it is therefore important to understand how these subjective biases operate. We define and explore a set of stylized tasks that allow us to highlight the effects of these subjective biases in LLM evaluation. In the process, we find evidence for several key underlying principles. First, there are multiple subjective LLM biases at play, and to understand the overall evaluation we must analyze the relative strengths of these biases and their interactions. Second, different prompts -- even when they seem superficially similar -- can have the effect of making different subjective LLM biases more or less salient. And third, there are complex interactions between the LLM's initial idea-generation phase and the subsequent decisions that go into prioritizing and pruning these ideas.

Diffusion models promise iterative refinement, yet masked diffusion language models (MDMs) suffer from two shortcomings: (1) inability to revise tokens, and (2) sparse training signal. The root cause is structural: MDMs control information flow only through input masking, so unmasked positions cannot be supervised, and self-correction is never learned. We propose the Leave-One-Out Transformer (LOOT), which addresses both issues. With LOOT, MDMs predict the distribution over potential replacements conditioned on the rest of the sequence at each position, in a single forward pass. Supervision can then be applied at masked and unmasked positions alike, yielding a lower-variance training objective with a dense loss signal at every token. Our method empowers MDMs to learn self-correction during training and supports continuous refinement at inference time via Gibbs sampling. LOOT achieves a new best validation perplexity on LM1B in a parameter-matched setting. Finetuned from a MDM checkpoint on OpenWebText, LOOT establishes a generative quality-vs-diversity frontier that surpasses prior remasking methods across NFE budgets.

Most energy-efficient federated learning (FL) methods report FLOPs or parameter reductions as evidence of energy savings. However, real GPU energy consumption differs across model architectures in ways that these computational proxies do not capture. In this work, we conduct an empirical study on reducing real GPU energy consumption in model-heterogeneous FL, where clients maintain distinct architectures and coordinate solely through shared prototypes. Through systematic experimentation with hardware-level energy measurements, we propose and validate two complementary mechanisms to make model-heterogeneous FL efficient in terms of real energy. First, Prototype-Aware Structured Pruning (PASP) evaluates each channel by the product of its batch-normalization scale magnitude and its gradient with respect to the prototype alignment loss---rather than the task loss---preserving channels critical for cross-client feature alignment while physically removing the rest. Second, Energy-Aware Training Scheduling (EATS) adaptively allocates local epochs to each client based on epoch-wise prototype alignment quality per unit energy, using smoothed alignment--energy profiles with a knee-point cutoff to prevent energy waste on epochs that yield diminishing alignment gains or amplify local drift. We investigate each mechanism's impact on accuracy and real energy consumption, as well as explore the impact of using both mechanisms simultaneously. All energy figures are obtained from GPU power monitoring rather than computational proxies, providing an empirically grounded evaluation of energy efficiency in heterogeneous federations.

Large language models are increasingly used to simulate human participants in social and behavioral studies, yet static persona prompting typically maps a participant profile and an experimental scenario directly to a response, entangling stable dispositions with situation-specific interpretations. To address this limitation, we introduce $\textbf{SPIN}$, a cognitive-affective personality system-inspired inference pipeline for social judgment and decision alignment. Specifically, SPIN implements this structured inference process through three zero-shot LLM calls that compile a task-blind participant core, elicit condition-specific cognitive-affective states, and read out decisions from those states, thereby reusing stable personality structure while routing each trial-specific response through an explicit state representation. We evaluate SPIN on two reconstructed social-psychological study families spanning uncertainty reasoning and pluralistic ignorance, across four base LLMs. Compared with blank, demographic, narrative, and chain-of-thought prompt variants, SPIN consistently delivers the strongest overall alignment performance across base LLMs and study families. Ablations and state analyses further show that both personality compilation and structured state elicitation contribute to the gains, and that the elicited states shift interpretably across informational and normative conditions. These results suggest that structured personality-state inference can improve benchmark-level behavioral alignment beyond richer persona descriptions or generic multi-step reasoning.

As AI models evolve from text and vision to physical agents, embodied AI faces a fundamentally different attack surface. Prior attacks on embodied systems have largely focused on semantic or instruction-level manipulations, such as prompts, adversarial images, and action commands; by contrast, we study environment-only perturbations that require no access to the model or instructions. We reveal a new attack surface: the physical environment itself. An adversary who rearranges objects, without modifying instructions or accessing models, can cause hazardous consequences. In this work, we propose EnvTrap, a diagnostic pipeline that constructs paired safe, trap, and null-trap (benign but misleading) embodied scenarios. We demonstrate that environment-only perturbations raise hazardous-action rates to an average of 86.5\% across multiple vision-language-action (VLA) models, with similar vulnerability patterns confirmed in world models. On a consequence-prediction task, model accuracy remains near chance, while human evaluators succeed easily. We further propose a consequence-aware defense that reduces trap trigger rates by an average of 76.3\% across VLA models in simulation and by 61.7\% on physical robots. This vulnerability arises because current embodied models can recognize scene state but often fail to predict action consequences under altered layouts. Our findings establish environment integrity as a prerequisite for safe embodied AI deployment. Our code and data are available at https://anonymous.4open.science/r/Envtrap-1BA4/

Earth Observation (EO) forecasting aims to predict future Earth surface dynamics from satellite observations under changing meteorological conditions. In this paper, we view this task as a partially observed, weather-driven world modeling problem, in which weather acts as a conditioning signal, while forecasting remains uncertain due to sparse observations and unobserved land-surface states. However, existing methods do not fully capture this setting: deterministic models collapse uncertainty into a single future prediction, while diffusion-based methods typically treat weather variables as undifferentiated conditioning signals, and existing benchmarks focus mainly on reconstruction accuracy rather than whether forecasts respond correctly to changed weather forcing. We introduce EO-WM, a video diffusion transformer for multispectral EO forecasting. EO-WM incorporates a physically informed conditioning framework that represents meteorological forcing through a climatological baseline, weather anomalies, and cumulative physical stress signals. Specifically, it separates baseline and anomaly through distinct conditioning pathways, and accumulates anomalous forcing over time to capture sustained heat and drought stress. To evaluate weather-response behavior beyond standard metrics, we introduce two diagnostic benchmarks: an Extreme Summer Benchmark for severity-aware prediction of vegetation degradation under extreme weather, and a Seasonal Matched-Pair Benchmark for testing response fidelity under changed weather forcing. Experiments show that EO-WM reduces the error in predicted Normalized Difference Vegetation Index (NDVI) decline amplitude by a relative 5.63\% and improves directional hit rate by a relative 7.80\%, while remaining competitive on standard pixel-level metrics.

Modern systems are often expected to transfer across tasks that were not specified during training, leading to the question: what data facilitates generalization in new, unanticipated settings? One hypothesis is that data with more structural information could contain shared circuits and subprograms that could be recycled in a wider array of downstream settings. Epiplexity, a recently proposed measure of the structural information a compute-bounded learner can extract from data, provides a mechanism to reason about this relationship. In this paper, we show how to operationalize epiplexity as an online training signal for data selection and generation. For selection, we fit scaling laws to the training loss curves of natural data domains to predict the expected epiplexity gain as a function of training tokens, and use this signal to adaptively determine the sampling weights over domains during training. For synthetic data generation, we define a generator's reward as the change in learner epiplexity over a buffer of previously generated data and use REINFORCE policy gradients to guide the generator toward an epiplexity-maximizing distribution. In both cases, higher epiplexity predicts improved downstream performance on zero-shot and fine-tuning based tasks, supporting the hypothesis that data rich in structural information yields representations that transfer across domains.

Zero-shot mutation effect prediction is widely used to evaluate protein language models (PLMs). However, because existing benchmarks mix additive and non-additive mutations, and non-additive effects (epistasis) are sparse yet critical, it remains unclear whether these models truly capture thermodynamic residue coupling. As a result, strong overall performance may mask weaknesses in modeling multi-residue interactions. To make this distinction explicit, we introduce EpistasisBench, a thermodynamically stratified zero-shot benchmark constructed from large-scale stability measurements. By isolating significant and sign epistatic double mutations, it separates additive effects from thermodynamic coupling and defines two complementary tasks: ∆∆G prediction restricted to epistatic subsets, and a direct ∆∆Gint scoring task that examines non-additive structure in model probability space. When evaluated across more than 17 state-of-the-art PLMs, EpistasisBench reveals a consistent pattern. Performance declines under strict epistasis filtering, strong single-site predictors fail on coupled mutations, and standard masked language model scoring shows no variation in ∆∆Gint, thereby preventing detection of non-additive effects. Together, these findings indicate that the limitation lies in common zero-shot inference schemes rather than model capacity. To demonstrate that non-additive signal can in fact be recovered under a more suitable inference formulation, we introduce a simple joint marginal inference mechanism that restores non-additive signal and improves performance without fine-tuning. Overall, EpistasisBench provides a physically grounded test for multi-residue interactions and highlights a key gap in current zero-shot evaluation of protein language models. The dataset is available at https://www.openml.org/d/47245

We consider the problem of synthesizing Clifford quantum circuits for devices with all-to-all qubit connectivity. We approach this task as a reinforcement learning problem in which an agent learns to discover a sequence of elementary Clifford gates that reduces a given symplectic matrix representation of a Clifford circuit to the identity. This formulation permits a simple learning curriculum based on random walks from the identity. We introduce a novel neural network architecture that is equivariant to qubit relabelings of the symplectic matrix representation, and which is size-agnostic, allowing a single learned policy to be applied across different qubit counts without circuit splicing or network reparameterization. On six-qubit Clifford circuits, the largest regime for which optimal references are available, our agent finds circuits within one two-qubit gate of optimality in milliseconds per instance, and finds optimal circuits in 99.2\% of instances within seconds per instance. After continued training on ten-qubit instances, the agent scales to unseen Clifford tableaus with up to thirty qubits, including targets generated from circuits with over a thousand Clifford gates, where it achieves lower average two-qubit gate counts than Qiskit's Aaronson-Gottesman and greedy Clifford synthesizers.

Scaling Federated Learning (FL) to billion-parameter models forces a challenging trade-off between privacy, scalability, and model utility. Existing solutions often tackle these challenges in isolation, sacrificing accuracy, relying on costly cryptographic tools, or introducing communication and optimization inefficiencies that affect convergence. We introduce ERIS, an FL framework centered on Federated Shard Aggregation (FSA), a novel mechanism that partitions each client update into non-overlapping shards whose aggregation is distributed across multiple client-side aggregators. FSA removes the central aggregation bottleneck, limits the information visible to any single observer, and preserves the centralized FL update after reassembly. ERIS can further readily integrate Distributed Shifted Compression (DSC) to reduce transmitted payloads and exposed coordinates. We prove that ERIS preserves convergence under standard assumptions and bounds mutual information leakage by the observable fraction of each update, decreasing with the number of client-side aggregators, and with the compression level when DSC is enabled. Experiments across image and text tasks, including large language models, show that ERIS achieves FedAvg-level utility while substantially reducing communication bottlenecks and improving robustness to membership inference and reconstruction attacks, without relying on heavy cryptography or utility-degrading perturbations.

Offline goal-conditioned reinforcement learning remains challenging in stochastic long-horizon settings, where compounding value estimation errors hinder reliable goal reaching. While prior methods have sought to address this challenge through various approaches, a key challenge remains path mirage, where agents overcommit to spuriously successful trajectories in offline datasets induced by stochasticity, often leading to failure on difficult tasks. To address this issue, we propose Branch-aware Graph Planning (BGP), which captures stochastic branching structures by identifying high-variability points as graph nodes and designing goal-conditioned edges with segmented learning to avoid unreliable branches and reduce unnecessary stochasticity along paths. As a result, BGP enables reliable goal reaching even in highly stochastic long-horizon environments, and experiments on diverse OGBench tasks show that it substantially outperforms prior state-of-the-art offline goal-conditioned RL methods.

Test-time model evolution offers a promising way for deployed models to improve from unlabeled test-time experience, yet most existing methods depend on backpropagation (BP), which incurs substantial memory overhead and makes them difficult to deploy on edge devices, quantized models, specialized accelerators, or black-box models. In this work, we study test-time model evolution under a strict two-forward budget, a setting that pushes adaptation toward highly efficient real-world deployment. We reveal three key obstacles in zeroth-order test-time optimization: susceptibility to shortcut solutions, uncontrolled weight drift, and ineffective update direction estimation. To overcome them, we propose EVA-0, a minimal zeroth-order adaptation framework that: 1) keeps the loss scale-invariant to prevent shortcut solutions; 2) devises an anchor-guided optimization strategy to alleviate weight drift; 3) uses sample-wise symmetric two-sided perturbation for update direction estimation and inference. EVA-0 requires no BP and performs both inference and adaptation within only two forward passes per sample. Results on ImageNet-C\&ViT-Base show that EVA-0 outperforms both BP-based DeYO and BP-free FOA, while achieving a 14$\times$ speed-up over FOA.

Systematic literature reviews (SLRs) are a demanding and high-stakes form of scientific knowledge synthesis that remains underspecified as an evaluation setting for large language models (LLMs). We introduce AgentSLR, a large-scale evaluation harness comprising an SLR automation workflow and an expert annotated dataset covering 16,248 articles, designed to test LLM capabilities across the stages of SLRs in epidemiology. Reference annotations were derived from peer-reviewed studies on WHO priority pathogens and produced by domain experts. AgentSLR evaluates each review stage as a separate unit with dedicated metrics enabling targeted failure analysis. We evaluated five frontier reasoning models and found that no single model dominated across all tasks, showing sub-task specialisation often hidden by aggregate benchmarks. Structured data extraction is a major bottleneck, with no model exceeding an average field-level F1 of 0.67. Estimated costs vary substantially, by up to 96 times across evaluated models. Documented failure modes suggest that the evaluated models are not yet reliable enough for unsupervised deployment in epidemiology, where findings can inform public policy.

Recent multimodal large language models (MLLMs) achieve strong results on video question answering, yet often fail to recover the temporal structure that makes a video coherent. They may merge neighboring events, miss single-event progression, or infer implausible relations across events. We study structured temporal understanding as the ability to recover event structure from visual evidence, including event boundaries, single-event progression, and multi-event relations. We argue that this gap arises from a mismatch between current post-training objectives and temporal reasoning, where caption-style supervision allows models to rely on language shortcuts rather than visual temporal grounding. To address this, we propose $\textbf{EventLens}$, a vision-centric reinforcement learning framework that learns video understanding through event-structure recovery}. Instead of reconstructing captions, EventLens trains models to recover temporal structure under controlled perturbations, such as altered temporal granularity, reversed progression, and shuffled event order. We instantiate this framework with a three-level temporal ontology and derive three verifiable task families: event segmentation, progression discrimination, and multi-event relation reasoning. These tasks admit deterministic rewards, enabling task-specialized GRPO training without human preference labels or learned reward models. We further introduce multi-teacher on-policy distillation (MT-OPD) to consolidate specialized policies into a unified model. Experiments on Qwen3-VL backbones show consistent improvements on temporal grounding and event-sensitive reasoning benchmarks, while preserving general video-QA performance. Learning curves further demonstrate that EventLens achieves comparable downstream transfer with significantly reduced training cost compared to direct mixed RL optimization.

Vision-Language-Action (VLA) models have emerged as a promising paradigm for robotic manipulation by unifying perception, language grounding, and action generation. However, they often struggle in scenarios requiring precise spatial understanding, as current VLA models primarily rely on 2D visual representations that lack depth information and detailed spatial relationships. While recent approaches incorporate explicit 3D inputs such as depth maps or point clouds to address this issue, they often increase system complexity, require additional sensors, and remain vulnerable to sensing noise and reconstruction errors. Another line of work explores implicit 3D-aware spatial modeling directly from RGB observations without extra sensors, but it often relies on large geometry foundation models, resulting in higher training and deployment costs. To address these challenges, we propose \textbf{Evo-Depth}, a lightweight depth-enhanced VLA framework that enhances spatially grounded manipulation without relying on additional sensing hardware or compromising deployment efficiency. Evo-Depth employs a lightweight \textit{Implicit Depth Encoding Module} (IDEM) to extract compact depth features from multi-view RGB images. These features are incorporated into vision-language representations through a \textit{Spatial Enhancement Module} (SEM) via depth-aware modulation, enabling efficient spatial-semantic enhancement. A \textit{Progressive Alignment Training} strategy is further introduced to align the resulting depth-enhanced representations with downstream action learning. Extensive experiments in simulation and real-world settings demonstrate the effectiveness of Evo-Depth. With only \textbf{0.9B} parameters, Evo-Depth achieves state-of-the-art-level performance across four simulation benchmarks. In real-world experiments, Evo-Depth attains the highest average success rate while also exhibiting the smallest model size, lowest GPU memory usage, and highest inference frequency among compared methods. These results demonstrate that lightweight implicit depth enhancement is an effective and practical solution for spatially grounded robotic manipulation. We release code and models to facilitate future research on lightweight depth-enhanced VLA models.

Human experts can accumulate experience from open-ended and complex inspection tasks, thereby continually improving their inspection capabilities. Actually, industrial hardware inspection faces challenges including open-set defect categories, complex defects (tiny, high-gloss, low-light, or blurred), high inference cost, and difficulty in accumulating experience post-deployment. Existing end-to-end inspection methods typically couple image restoration with defect recognition in a single forward pass, limiting generalization to unseen components or composite defects; cascade “restore-then-detect” approaches often rely on static preprocessing, hindering adaptive adjustment based on scenario, defect morphology, and historical failures. To address this, we propose EvoInspect, the first memory-enhanced multi-modal multi-agent framework for industrial defect object detection. It leverages a multi-modal agentic memory mechanism to distill self-evolving scenario-specific Defect Detail Restoration decisions and dynamically grounded trajectories, enabling “accumulating experience like human experts.” Specifically, (1) we introduce a dynamic Defect Detail Restoration strategy, including adaptive slicing for tiny defects, highlight suppression, low-light enhancement, super-resolution, and binarization, combined with a coarse-to-fine token-efficient grounding mechanism for efficient defect capture; (2) we design a dual-channel distillation mechanism unifying restoration and detection (perception and execution) experience, incorporating both successful and failed validation trajectories to support system self-evolution. Experiments on bearing surfaces, PCBs, solar electroluminescence (solar EL), and magnetic tile datasets show that EvoInspect consistently outperforms strong baselines in defect localization recall, inference accuracy, while demonstrating cross-scenario adaptability and offline continual improvement.

Building agentic systems that can autonomously self-improve from experience is a longstanding goal of AI. Large language models (LLMs) today primarily self-improve via two mechanisms: self-reflection for context updates, and reinforcement learning (RL) for weight updates. In this work, we propose Evolutionary System Prompt Learning (E-SPL), a method for jointly improving model contexts and model weights. In each RL iteration, E-SPL samples trajectories under multiple system prompts in parallel, then jointly applies RL updates to weights and evolutionary updates to system prompts via LLM self-reflection. E-SPL encourages a natural division between declarative knowledge encoded in prompts and procedural knowledge encoded in weights. Most notably, in an easy-to-hard generalization setting (AIME $\rightarrow$ BeyondAIME), E-SPL improves RL success rate from 38.8% $\rightarrow$ 45.1%. E-SPL also improves RL on AIME 2025 (56.3% $\rightarrow$ 60.6%), HMMT 2025 (50.0% $\rightarrow$ 52.7%), and agentic search (44.2% $\rightarrow$ 48.6% on gpt-oss-120b). Across all settings, E-SPL outperforms both RL-only and evolution-only baselines, demonstrating that weight updates and context updates are deeply synergistic and can together yield gains in generalization that neither achieves alone.

Automating optimization modeling from natural language faces two key challenges: training corpora lack structural diversity, and data generation pipelines remain static and decoupled from model learning. To address these challenges, we propose EvoOptiGraph, a novel framework where data and model co-evolve, driven by model weaknesses. EvoOptiGraph represents each mixed-integer linear program (MILP) as an attributed bipartite graph and applies validity-preserving evolutionary operators to generate structurally diverse instances. The evolved graphs are converted into solver code and natural language via deterministic compilation and verified back translation. Training proceeds in two stages: supervised fine-tuning (SFT) on an initial dataset, followed by reinforcement learning with verifiable rewards (RLVR), where graph-derived weakness signals dynamically evolve new instances that target the model's failures, forming a closed loop that continuously updates the training distribution. Empirical results on six public datasets show that EvoOptiGraph significantly outperforms larger generalist models, agentic methods, and specialized baselines in accuracy, executability, and generalization. These results demonstrate that targeted data–model coevolution is an effective strategy for improving LLMs on optimization modeling tasks.

Current evaluations of LLM safety predominantly rely on *severity-based taxonomies* to assess the harmfulness of models' responses to malicious queries. We argue that this formulation requires re-examination as it implicitly equates model output with realized harm while neglecting *Execution Likelihood*—the conditional probability of a threat being realized by a real-world actor given a model's response. In this work, we introduce **Expected Harm**, a metric that weights the severity of a jailbreak by its execution likelihood, modeled as a function of *execution cost*. Through empirical analysis of state-of-the-art models, we reveal a systematic *Inverse Risk Calibration*: models disproportionately exhibit stronger refusal behaviors for low-likelihood (high-cost) threats while remaining vulnerable to high-likelihood (low-cost) queries. Mechanistically, we first provide an analysis using linear probing, revealing that while refusal mechanisms linearly separate queries by severity, they do not use execution likelihood as an indicator for refusal. Then, we use Olmo-3's open-source staged checkpoints to trace the root cause of miscalibration to the Supervised Fine Tuning (SFT) training data distribution (82.8\% of safety examples at cost levels 0–1) and show that cost representations are weak from the SFT stage onward. Finally, we demonstrate that this miscalibration creates a vulnerability: by exploiting this property, we increase the attack success rate of existing jailbreaks by up to $2\times$.

Standard In-Context Reinforcement Learning (ICRL) typically assumes stationary dynamics within an interaction history, a restriction that limits its real-world applicability. We formalize Lifelong ICRL, a more realistic paradigm in which agents must continuously adapt to diverse non-stationary dynamics, ranging from abrupt random shifts to structured temporal drifts within a single lifetime. To probe the adaptation limits of this setting, we conduct a large-scale empirical study across non-stationary environments that span discrete symbolic reasoning and continuous physics-based control. We systematically investigate three core dimensions governing adaptation: (i) Training Regimes, comparing the generalization boundaries of domain randomization, stationary training, non-stationary training, and their staged combinatorial strategies; (ii) Model Architectures, evaluating modern sequence models, including Linear Attention and hybrid designs; and (iii) Impact of Scale, analyzing the relative contributions of interaction scale and model capacity. Our results reveal how these factors jointly shape robust in-context adaptation under unknown dynamics, yielding actionable insights for the design of generalist agents.

We aim to design revenue-maximizing single-item auctions that are deterministic, strategy-proof, and ex post individually rational --- the quintessential fundamental model for optimal mechanism design. While Myerson's seminal work solved this model for independent bidders, straightforward extensions to correlated settings often result in non-monotonic, and therefore non-strategy-proof allocations. Critically, optimal allocation design for correlated bidders is NP-hard to approximate, rendering theoretical optimality guarantees in this domain computationally intractable. We establish a new standard of empirical rigor for this theoretically hard model by proposing an empirical pipeline. We synthesize neural interpolation of Myerson's greedy allocation based on marginal profits (the correlated variant of virtual valuation), formal neural verification to enforce exact strategy-proofness, and a constructive repair procedure. Empirically, our method is highly effective, consistently achieves near-optimal revenue across a wide range of distributions, including synthetic, adversarial, and real-world (consumer, financial, and industrial) datasets. Compared to existing manual and neural baselines, our approach shows substantial improvement, often reducing the revenue gap by an order of magnitude. Our study is also the first to introduce (unattainable) greedy revenue as a rigorous upper bound for empirical benchmarking, providing a definitive quantitative performance measure where true optima remain theoretically elusive. We demonstrate the generality of our approach by extending it to multi-unit auctions with unit demand and integrating our verification techniques into RegretNet to achieve exact strategy-proofness.

Electrocardiogram (ECG) foundation models pretrained on typical diagnostic 10-second ECG segments, have demonstrated strong transferability across a range of clinical applications. However, many real-world applications produce recordings that are typically longer, and are varied in duration during inference time. These 10-second models have no built-in way to combine information across time. Extending them to longer horizons introduces two challenges: structural incompatibilities arising from input-length disparities, and semantic challenges that limit meaningful temporal aggregation. We propose a parameter-efficient framework that extends pretrained ECG foundation models to longer and variable-length ECGs without retraining the backbone. Guided by a frozen pretrained 10-second model, we introduce a lightweight plug-in module that extends the model in two complementary ways: (i) structurally compatible long-sequence processing and (ii) semantically informed temporal modeling. Experiments on multiple long-horizon ECG tasks, datasets, and foundation model backbones demonstrate that our method enables robust long-horizon extension from pretrained snapshot models, consistently outperforming sliding-window and pooling-based baselines with strong parameter efficiency.

Large language models (LLMs), especially reasoning models, generate extended chain-of-thought (CoT) reasoning that often contains explicit deliberation over future outcomes. Yet whether this deliberation constitutes genuine planning, how it is structured, and what aspects of it drive performance remain poorly understood. In this work, we introduce a new method to characterize LLM planning by extracting and quantifying search trees from reasoning traces during four-in-a-row gameplay. By fitting computational cognitive models on the extracted search trees, we characterize how plans are structured and how they influence move decisions. We find that LLMs' search is shallower than humans', and that performance is predicted by search breadth rather than depth. Most strikingly, although LLMs expand deep nodes in their traces, their move choices are best explained by a myopic model that ignores those nodes entirely. A causal intervention study in which we selectively prune CoT paragraphs further suggests that move selection is driven predominantly by shallow rather than deep search. These patterns contrast with human planning, where performance is driven primarily by deep search. Together, our findings reveal a key difference between LLM and human planning: while human expertise is driven by deeper search, LLMs do not act on deep lookahead. This dissociation offers targeted guidance for aligning LLM and human planning. More broadly, our framework provides a generalizable approach for interpreting the structure of LLM planning across strategic domains.

Visual Autoregressive (VAR) models have emerged as a strong alternative to diffusion for image synthesis, yet their fixed training resolution prevents direct generation at higher resolutions. Naively transferring training-free extrapolation methods from LLMs or diffusion models to VAR yields three characteristic failure modes: global repetition, local repetition, and detail degradation. We trace them to a unified band-stage mismatch: VAR generates images in a coarse-to-fine, scale-wise process where each stage is driven by a distinct dominant RoPE frequency band, and each failure mode emerges when the dominant band of a particular stage is disrupted. Building on this insight, we propose Stage-Aware RoPE Remapping, a training-free strategy that assigns each frequency band a stage-specific remapping rule, jointly suppressing all three failure modes. We further observe that attention becomes systematically dispersed as the image resolution increases. Existing methods typically depend on predefined attention scaling factors, which are neither adaptive to the target resolution nor capable of faithfully capturing the actual extent of attention dispersion. We therefore propose Entropy-Driven Adaptive Attention Calibration, which quantifies dispersion via a resolution-invariant normalized entropy and yields a closed-form per-head scaling factor that realigns the extrapolated-resolution attention entropy with its training-resolution counterpart. Extensive experiments show that our method consistently outperforms prior resolution-extrapolation methods in both structural coherence and fine-detail fidelity. Our codes have been released in supplementary materials and will be released on Github.

Editing a local segment in a talking head video without reshooting the entire scene remains challenging and underexplored. Given local edits, such as insertion, deletion, or substitution, talking head video editing must synthesize a replacement segment while leaving the rest unchanged. This requires variable-duration local rewriting while preserving identity, unedited regions, and boundary continuity, which standard speech-driven generation and lip synchronization do not directly address. Moreover, direct supervision is infeasible, as it requires paired videos of the same person and scene differing only in a local spoken segment, which does not exist in the real world. We instead formulate talking head video editing as facial motion infilling, a self-supervised pretext task that recovers masked facial motion from speech and surrounding motion context in ordinary video--speech pairs. The key insight is that local video editing can be simulated during training by masking a motion span and reconstructing it from speech and visible motion context. Based on this formulation, we introduce FacEDiT, a mask-controlled talking head model with local temporal attention bias and temporal smoothness regularization for improved lip--speech alignment and transition continuity. We also introduce FacEDiTBench, the first benchmark for talking head video editing, covering diverse edit types and lengths with dedicated evaluation metrics. Extensive experiments show that FacEDiT produces accurate, speech-aligned edits with strong identity preservation and seamless boundary transitions. Beyond editing, the same facial motion infilling model extends to portrait animation and lip synchronization by simply changing the mask pattern, establishing FacEDiT as a unified framework for talking head video editing and generation. \textit{We will release the code and data upon acceptance.

Three-dimensional (3D) anomaly detection underpins quality assurance in advanced manufacturing and engineering, capturing subtle defects and deviations that 2D inspection struggles to resolve due to occlusion, viewpoint, and appearance confounds. Existing 3D anomaly detection methods either rely on reconstruction errors or stored normal representations to detect anomalies. While both paradigms have driven substantial progress, they share a fundamental limitation: both operate entirely in the visual domain, missing the rich semantics encoded in natural language, which also typically constrains them to per-category models. We propose FACETS, among the first frameworks that leverage a 3D vision--language model for 3D anomaly detection, opening a new paradigm beyond reconstruction- and memory-bank-based methods. The key idea is to explicitly retain native point-level features that enable reasoning at both patch and point granularities, and further leverage language grounding and cross-granularity geometric modeling along two complementary axes, linguistic semantics and geometric saliency, so that coarse semantic cues and fine-grained geometric details jointly support anomaly detection. FACETS enables unified multi-category anomaly detection, avoiding the per-category models used by most prior methods. Notably, we provide a mathematical analysis of our loss function, offering valuable insights into the substantial improvement FACETS achieves in anomaly localization. Extensive experiments on popular benchmarks reveal that FACETS establishes new SOTA performance across datasets and metrics for 3D anomaly detection, substantially and consistently outperforming existing methods. Code is provided as supplementary material.

Fairness has emerged as a central consideration in machine learning, motivating the study of fair range $k$-supplier clustering as a fundamental problem that focuses fairness on the selected centers. Given a set of suppliers and clients, where each supplier may belong to one or more demographic or functional groups, the objective is to select $k$ suppliers that minimize the maximum client-to-center distance while ensuring that the number of selected suppliers from each group satisfies prescribed lower and upper bounds. For disjoint supplier groups, we develop a polynomial-time $3$-approximation algorithm in the offline setting and a $(3+\epsilon)$-approximation algorithm in the streaming setting. We further consider overlapping supplier groups and show that this generalization admits a parameterized $3$-approximation algorithm whose runtime is exponential in the number of clusters. Finally, experiments on both synthetic and real-world datasets demonstrate that our algorithms achieve significantly better clustering quality and runtime efficiency in comparison with state-of-the-art methods.

Objectives involving bilinear forms (u^\top f(A(\theta))v) for Hermitian (A) arise widely in scientific computing and probabilistic machine learning. For large matrices, Lanczos efficiently approximates these quantities, but differentiating them with respect to (\theta) is challenging. Existing approaches either backpropagate through the Lanczos recurrence, requiring reorthogonalization for stability, or apply Arnoldi to an augmented block matrix of twice the original size. Both introduce extra computation and orthogonalization costs that can limit performance on modern hardware. We propose a forward-only gradient approximation that reuses the Lanczos pass and adds very minimal overhead in most cases. We prove that its error is proportional to the Lanczos residual norm, the same quantity controlling the forward approximation. Whereas a traditional adjoint-based calculation would be unstable without reorthogonalization, the new method appears unconditionally stable in our tests. It is also faster than existing state-of-the-art approaches.

Light-field (LF) spatial super-resolution hinges on exploiting cross-view correspondences to recover high-frequency details while preserving geometric structure. However, existing methods often rely on implicit feature interaction or coarse disparity maps, which suffers noticeable performance degradation in regions with large disparities. In this paper, we propose Fine-Grained Epipolar Attention Diffusion (FEAD), which adopts epipolar attention as an explicit mechanism to characterize cross-view geometric correspondence along epipolar lines. To enable more accurate feature alignment and long-range spatial–angular interaction, we introduce a diffusion-based epipolar attention refinement strategy that progressively improves the initial attention. With the refined attention as guidance, cross-view features are explicitly warped and fused to enforce geometric consistency and aggregate complementary details across views for fine-grained reconstruction. Extensive experiments demonstrate that FEAD achieves state-of-the-art performance, with particularly strong gains in large-disparity scenarios.

Recent large-scale outages at major cloud providers such as AWS and GCP have exposed the fragility of relying on a single cloud. To improve resilience, fault tolerance, and business continuity, many enterprises are moving their services to multi-cloud environments. However, multi-cloud deployment also makes workload forecasting substantially harder. Existing multi-cloud workload forecasting methods are typically designed for either centralized or isolated environments, leading to risks of private data leakage or limited generalization. To address these challenges, we propose \textbf{FedCAG}, a \textbf{Fed}erated \textbf{C}ausality-\textbf{A}ware \textbf{G}raph learning paradigm for multi-cloud workload forecasting. Instead of treating federated learning as parameter averaging over local predictors, FedCAG jointly federates predictive models and graph-structured dependency priors. Each client constructs causality-aware, spatial, and temporal graphs from its private telemetry to model directed inter-metric influence, metric-level interactions, and intra-window temporal dynamics. To address strong cross-cloud heterogeneity, FedCAG further combines causality-aware representation fusion, adaptive graph refinement, and client-specific personalization, enabling the global model to benefit from shared workload structures while preserving local specificity. Experiments on real-world datasets show that FedCAG consistently outperforms mainstream federated baselines and even several centralized methods trained on the full dataset, delivering stable and accurate forecasting across heterogeneous and privacy-constrained deployments. The source code is available at: \url{https://anonymous.4open.science/r/FedCAG-FB9C}.

Federated Graph Learning (FGL) aims to collaboratively train Graph Neural Networks (GNNs) across distributed subgraphs with a primary challenge of missing cross-client connections. Existing approaches predominantly rely on a retrieval-based paradigm, which reconstructs missing context by accessing external information across clients. However, such cross-client transmissions inevitably expand the attack surface for privacy inference and incur significant communication overhead. In this work, we propose \textbf{FedLMC}, a novel framework that eliminates the necessity of any node information sharing and prior global adjacency knowledge. FedLMC exploits a dual-stage message compensation mechanism to generate informative and expressive node representations. For the semantic information loss of external neighbors, we propose Local Semantic Compensation, which exploits semantic expansion via semantic surrogates to compensate for missing semantic context. For the structural information loss induced by semantic expansion, we propose Community Structural Compensation, which exploits structural anchoring via community anchors to compensate for underlying structural context. Extensive experiments on twelve datasets (both homophilic and heterophilic) demonstrate that FedLMC establishes a new state-of-the-art performance with superior generalization and robustness to highly fragmented graph data.

Federating a LGN is challenging because each gate is parameterised by a distribution over 16 Boolean operations, so averaging parameters across clients does not correspond to averaging the underlying functions. Under non-IID data, such parameter-space merging collapses to near-chance accuracy. Boolean Feature Selection (BFS) avoids this issue by selecting discrete Boolean functions rather than averaging parameters. We propose Federated Boolean Feature Selection (FBFS), which extends BFS to LGNs by treating hardened last-layer gates as candidate Boolean features and aggregating per-class, per-gate sufficient statistics across clients. Clients share only firing aggregates computed on a held-out split, allowing the server to reconstruct discriminative scores, defined as the gap between a gate's mean firing for a class and for the remaining classes, without accessing raw data. This procedure is equivalent to centralized BFS on the union of client data. We further introduce a diversity-regularized variant of FBFS that encourages complementary feature selection across gates. Empirically, FBFS is the only data-free method we evaluate whose performance remains stable under extreme non-IID settings, outperforming parameter-space merging and other baselines by large margins. It maintains high accuracy across a wide range of heterogeneity levels and scales to large architectures. In addition, FBFS enables efficient deployment: the resulting models compile to hardware-efficient logic representations with substantially fewer post-synthesis cells than ensemble-based alternatives while remaining bit-exact with their software counterparts.

World models aim to improve robotic decision making by predicting the consequences of actions. However, in practice, their predictions often become unreliable once the robot encounters states outside the training distribution, limiting their effectiveness at deployment. We observe that execution itself provides a natural but underutilized signal: after each action, the robot directly observes the true next state, revealing the mismatch between predicted and actual outcomes. Building on this insight, we propose $\textbf{feedback world model}$, a new paradigm that closes the loop between prediction and observation at inference time. Instead of treating the world model as a static open-loop predictor, our method maintains a lightweight feedback state that is updated online to iteratively correct future predictions, compensating for model errors using real-time observations without additional training data or parameter updates. We show that this process can be interpreted as a latent-space observer and admits convergence guarantees under mild conditions. We further introduce action-aware guidance to better translate corrected predictions into control by emphasizing action-controllable components while suppressing irrelevant variations. Experiments on LIBERO-Plus, Robomimic, and real-world manipulation tasks demonstrate that our method substantially improves both prediction accuracy and policy performance under distribution shift. In particular, it reduces world model prediction error by up to 76.4\% and improves out-of-distribution (OOD) success rate by 30\%. These results show that incorporating real-time feedback at inference time provides a simple yet powerful alternative to static world modeling.

Discovery of novel materials for physical memory is essential for superior memory systems enabling AI hardware scaling. Among them, ferroelectrics hold promise for highly scalable DRAM. Yet their discovery remains driven largely by human intuition and exhaustive database screening. We present Ferrogen, a generative pipeline for the targeted discovery of novel ferroelectric materials for memory by fine-tuning the diffusion-based crystal structure generator Mattergen on ferroelectric-relevant properties. To enable property-conditioned generation, we construct a ML-labeled dataset using a suite of fast machine learning estimators, including a novel two-stage ensemble polarization predictor, dramatically reducing reliance on high-throughput Density Field Theory (DFT) during training and screening. The base Mattergen model is fine-tuned via lightweight adapter layers conditioned on polarization, switching energy, and metal-probability embeddings, with classifier-free guidance enabling targeted generation of candidates with high polarization, low but finite switching barriers, and insulating character. Generated candidates are screened by the same ML estimators and validated through rigorous DFT calculations, including structural relaxation, band gap verification, and Berry phase polarization computation. Against the ML screens, Ferrogen achieves a roughly $80 \times$ improvement in search efficiency over exhaustive database search. The screened candidates are then finally validated using DFT and made publicly available as a database of theoretically confirmed novel ferroelectrics. The pipeline is readily extensible to other functional material classes, establishing generative models as a powerful paradigm for accelerating electronic materials discovery for high performance memory.

LiDAR-based 3D Object Detection (L3OD) is a fundamental 3D perception task, yet its performance is often compromised by domain shifts arising from diverse sensor configurations and environmental conditions. Existing Test-Time Adaptation (TTA) methods primarily leverage self-supervision in the spatial domain, yet they frequently confound semantic content with domain style due to the sparse, non-Euclidean nature of point clouds. To overcome this limitation, we propose FIND (Frequency INvariance Disentanglement), a novel TTA framework that precisely disentangles domain-invariant content from domain-specific interference within the frequency domain. Our methodology utilizes B-Spline fitting to construct locally adaptive filters integrated into a dual-stream architecture: an Invariant Stream extracts robust features to guide pseudo-labeling, while a Specific Stream enforces consistency against domain perturbations. By combining invariance-driven alignment with stability-guided regularization, our approach dynamically extracts robust domain-invariant features while suppressing domain-specific interference. Extensive experiments on cross-dataset adaptation and robust corruption benchmarks demonstrate that FIND significantly outperforms state-of-the-art methods.

Evaluation of foundation models often rely on aggregate scores from benchmarks that lack comprehensive coverage and metadata for a fine-grained evaluation. We introduce a framework for automated benchmark generation. Our framework generates evaluation problems grounded in reference material, such as textbooks, producing benchmarks with broad coverage, rich metadata, and robustness to contamination. The pipeline employs a multi-agent architecture for problem generation and a solution-graph-driven strategy that significantly improves the reliability of ground truth solutions. Using the framework, we generate three benchmarks in Machine Learning, Corporate Finance, and Personal Finance. Expert review finds a significantly lower ground-truth error rate than previous benchmarks such as MMLU and GSM8K. Evaluation of 12 commercial and open-source models shows that our benchmarks achieve near-uniform competency coverage and surface performance differences across models that existing benchmarks fail to capture. We open-source the framework and our curated benchmarks.

A fruitful approach towards understanding generalization in machine learning involves characterizing how models separate generalizable structure from idiosyncratic or even corrupted features of training data. Fisher information has emerged as a diagnostic tool for this separation, leveraging loss geometry to distinguish generalization from memorization. Here we investigate this connection in vision models trained on CIFAR-10 with controlled label noise that can be learned only through memorization. We show that projecting model weights onto the top Fisher eigenvectors decouples general task performance from noisy sample memorization, and we identify the layer in which this decoupling emerges. We show that the same phenomena are present in small transformers trained to perform in-context learning and reproduced in closed form in a simple, analytically solvable linear model. Our results reveal that Fisher information captures aspects of the geometric structure underlying how neural networks allocate capacity between generalization and memorization.

Designing functional biological sequences requires navigating vast discrete spaces under strict evolutionary and biophysical constraints. Discrete Flow Matching (DFM) offers a generative framework over such spaces, but existing approaches rely on biologically uninformative couplings and offer limited flexibility for variable-length sequence generation and fine-grained control. We propose a structured coupling that encodes domain-specific preferences among sequence elements, biasing the source distribution toward plausible regions without modifying the flow objective or training procedure. Building on this, we introduce a latent edit-based rate parameterization that models variable-length generation via edit operations conditioned on a shared global latent, akin to a latent variable model, while remaining tractable. We further introduce a latent classifier-free guidance mechanism that steers generation coherently in continuous latent space, along with Dirichlet-prior temperature scaling for test-time control over edit operations. Our method achieves state-of-the-art performance across diverse biological sequence tasks, including density estimation, unconditional and conditional DNA sequence generation, and peptide sequence generation.

A key strategy for balancing performance and cost in modern machine learning systems is to dynamically route queries to either a low-cost model or a more expensive oracle (such as a large pretrained model or human expert), an approach known as model routing. In this work we present a new uncertainty-aware router that (1) avoids unnecessary oracle calls on inherently ambiguous queries, and (2) adapts dynamically to different loss functions and cost parameters through simple hyperparameter changes, without retraining. Our method, applicable to any classification setting where multiple independent annotations per input are available, is based on decomposing total uncertainty into irreducible and reducible components using higher-order predictors [Ahdritz et al., 2025]. This enables a unified approach to both routing and abstention: predict with the weak model when uncertainty is low, route to the oracle when reducible uncertainty is high, and abstain when irreducible uncertainty is high. Our router comes with strong theoretical guarantees bounding regret relative to optimal task-specific routers. We conduct experiments on both synthetic and real-world datasets that demonstrate the benefits of our approach in suitable regimes---in particular, whenever reducible and irreducible uncertainty are not too correlated.

Large language models (LLMs) are increasingly deployed in sensitive settings such as software engineering, where their outputs directly shape downstream artifacts. Recent work has shown that an identical model can produce measurably different outputs depending on the deployment platform, a consequence of non-associative floating-point arithmetic and divergent kernel implementations. We study the security implications of this platform-dependent variability and uncover a novel attack surface on LLM deployments. We introduce FloatDoor, the first input-independent, platform-triggered backdoor attack against generative LLMs. The compromised model exhibits adversary-chosen behavior when served on a target platform and is otherwise benign. FloatDoor is realized through two lightweight LoRA adapters, one that amplifies inter-platform numerical divergence and one that binds the resulting platform signature to a malicious downstream task, while leaving aggregate model utility largely intact. FloatDoor exploits a pronounced time-of-check, time-of-use gap between model auditing and serving. We demonstrate FloatDoor on Qwen3-4B across a broad range of deployment targets, including NVIDIA GPUs, Google TPUs, AWS Graviton, and Alibaba Yitian-710. As a final case study, we show that FloatDoor reliably induces exploitable code vulnerabilities on a chosen target platform. Our results establish a new class of attacks on LLM deployments and underscore the pressing need for trusted model supply chains in sensitive, LLM-powered applications.

Reference-based image quality assessment (IQA) metrics aim to reflect how humans perceive the perceptual distance between a pair of images. To learn how the human visual system (HVS) operates, recent reference-based IQA metrics heavily rely on human-annotated data. Mean opinion score (MOS)-based pointwise scoring, which assigns a scalar quality value per image, is preferable for annotation but is prohibitively expensive to collect at scale and is known to be noisy due to inconsistent human judgments. As an alternative, two-alternative forced choice (2AFC) pairwise labels have gained popularity due to their reliability and efficiency, but they capture only relative comparisons between pairs. In this paper, we propose a fully automated data generation pipeline that generates pointwise perceptual distance labels between image pairs without any human annotation. Our approach exploits the generative dynamics of diffusion models as a perceptual distance proxy, where the coarse structure of an image is generated in the early timesteps and the fine details are generated in the later timesteps. Images that fork early in the generation process share only coarse structure and are perceptually far apart; images that fork late differ only in fine detail. We demonstrate that the diffusion trajectory aligns well with the human visual system, and use this forking moment, FoMo, as a reference-grounded distance label to supervise the training of a reference-based IQA metric. The pointwise labels, which support universal comparison between arbitrary image pairs, enable an information-rich training objective. Extensive experiments across diverse backbone architectures confirm the effectiveness of our generation pipeline, outperforming human-annotated datasets in multiple benchmarks.

Label-supervised manifold alignment bridges the gap between unsupervised and correspondence-based paradigms by leveraging shared label information to align multimodal datasets. Still, most existing methods rely on Euclidean geometry to model intra-domain relationships. This approach can fail when features are only weakly related to the task of interest, leading to noisy, semantically misleading structure and degraded alignment quality. To address this limitation, we introduce FoSTA (Forest-guided Semantic Transport Alignment), a scalable alignment framework that leverages forest-induced geometry to denoise intra-domain structure and recover task-relevant manifolds prior to alignment. FoSTA builds semantic representations directly from label-informed forest affinities and aligns them via fast, hierarchical semantic transport, capturing meaningful cross-domain relationships. Extensive comparisons with established baselines demonstrate that FoSTA improves correspondence recovery and label transfer on synthetic benchmarks and delivers strong performance in practical single-cell applications, including batch correction and biological conservation.

Large language model (LLM) agents are highly vulnerable to memory-poisoning attacks. Existing defenses primarily rely on external modules for either static memory isolation or continuous online auditing, resulting in low memory efficiency and high computational overhead. To address these limitations, inspired by neuroscientific mechanisms that weak reactivation can destabilize memories and induce decay in the absence of reinforcement, we propose an agent-intrinsic training-free memory repair framework, \textbf{DREAM} (\textbf{D}ynamic \textbf{R}eactivation for \textbf{E}ngram \textbf{A}ttenuation in \textbf{M}emory), that enables selective functional forgetting of poisoned memories. Specifically, DREAM implements a three-stage pipeline: perturbation-induced implicit reactivation to reveal the structural fragility of malicious memories, adaptive anomaly diagnosis based on multi-dimensional activation patterns to detect topological anomalies, and a dynamic memory repair module that selectively suppresses harmful memories while reinforcing benign ones. Extensive experiments across diverse attackers and real-world tasks demonstrate that DREAM reduces attack success rates by over 95\% against backdoor poisoning while preserving strong benign utility. DREAM also achieves competitive robustness against injection attacks. In terms of runtime and token consumption, DREAM achieves up to a 2.13$\times$ speedup and a 37.7\% reduction in token consumption compared with A-MemGuard. Furthermore, DREAM also achieves a task success rate of 92.96\% on poisoned multi-agent systems.

Medical image segmentation requires the precise alignment of macroscopic semantics and microscopic geometry, yet existing representation paradigms struggle to balance massive low-frequency semantic context with ultra-sparse, high-frequency boundary details. To address this fundamental information asymmetry, we introduce Fractal-G, a plug-and-play multi-scale fusion module that presents a continuous Grid-Graph-Grid feature reconstruction paradigm. Within this framework, an Uncertainty-Aware Fission Router dynamically translates regular image grids into a non-uniform heterogeneous graph, adaptively allocating dense microscopic nodes to complex boundaries while retaining sparse macroscopic nodes in homogeneous background regions. To ensure unbiased feature aggregation across these varying physical scales, an Area-Aware Continuous Neighborhood Graph explicitly incorporates physical node sizes into the topological message passing. Finally, a Topology-Aware Implicit Renderer projects the unstructured graph back into dense, artifact-free continuous feature fields. Extensive experiments on diverse medical image segmentation tasks (including skin lesions and polyps) demonstrate that Fractal-G easily integrable with mainstream backbones, consistently achieving state-of-the-art geometric fidelity and cross-resolution robustness with only 12\% GFLOPs increase (U-Net based). Our code is publicly available at https://anonymous.4open.science/r/Fractal-G/.

State Space Models (SSMs) compress sequence history into a bounded recurrent state, making the resulting memory law a central architectural choice for long-context performance. Most modern SSMs rely on ODE-based dynamics that lead to exponential forgetting, limiting their ability to retain information over broad temporal ranges. We introduce FRAC, a selective SSM architecture derived from fractional dynamics that replaces this exponential decay with power-law long memory. To make fractional dynamics practical, FRAC approximates the heavy-tailed target kernel with a finite-state, log-spaced sum of exponential modes. This construction turns fractional memory into an efficient recurrent module with parallel training and prefill, while retaining bounded-state autoregressive decoding. Extensive experiments, including 1.3B-parameter language modeling, demonstrate that FRAC consistently improves long-context performance over state-of-the-art SSM baselines while staying competitive on short-context. These results show that fractional dynamics provide a practical and effective prior for long-context SSMs.

Long video understanding requires VideoLLMs to answer queries about videos spanning tens of minutes to hours, but such long videos produce far more visual tokens than current VideoLLMs can process. To address this issue, keyframe selectors are proposed to select the most query-relevant frames by computing the similarity between frame and query embeddings. Existing selectors either encode frames independently without temporal context or process chunks in isolation with ranking objectives that are difficult to optimize. To alleviate those limitations, we propose FrameScout with two core designs: (i) a streaming selector architecture that leverages a sliding-window KV cache and successor aggregation to produce temporally-contextualized frame embeddings; (ii) a frame-query contrastive objective that aligns frame and query embeddings in a shared space, directly training the selector to distinguish relevant frames from irrelevant ones. Extensive experiments on three long-video benchmarks and six base VideoLLMs demonstrate that FrameScout consistently achieves state-of-the-art performance.

Multimodal Recommender Systems (MRSs) have achieved significant success in personalized recommendations by integrating diverse modalities such as images, text, and audio. While in complex real-world industrial scenarios where MRS has usually been deployed, the presence of missing modalities is very common. Most existing MRS methods recover missing data by disentangling available modality features into general and specific representations, thereby generating missing modalities based on personalized aligned feature. However, these methods rely on implicit constraints within pre-extracted multimodal feature spaces, which usually introduce noise that prevents the effective decoupling of general and specific semantics, thereby diminishing recommendation accuracy. Moreover, unguided linear transformations of available modality features for generation lead to semantic hallucinations, which inevitably degrade the ranking quality of the recommendation results. To address these issues, we propose the Frequency-Decoupled Knowledge Distillation Framework for Multimodal Recommendation (\textbf{FreDRec}), which conducts effective and robust missing feature generation. Specifically, motivated by the physical characteristics of multimodal signals, FreDRec utilizes Fourier and Discrete Wavelet Transforms to explicitly decouple raw multimodal signals into low-frequency core semantics and high-frequency details, which is mathematically proven to effectively minimize noise introduction at the source. Next, to ensure robust feature generation, we pre-train a teacher model based on decoupled multi-view graph propagation to guide a student model via multi-level knowledge distillation, which establishes an end-to-end highly efficient lightweight architecture and achieves precise missing modality generation with expert collaborative bounds. Finally, extensive experiments on multiple public benchmark datasets demonstrate that FreDRec achieves superior performance under various modality missing ratios.

Adapting pretrained Vision-Language-Action (VLA) policies to new deployment environments typically requires collecting expert demonstrations in each target domain, which is costly and difficult to scale. In this paper, we present FreeAct, a framework for demonstration-free robot adaptation that converts generated videos into physically grounded supervision. However, a key challenge is that generated manipulation videos, while visually plausible, often contain embodiment artifacts and kinematic inconsistencies that make direct inverse-dynamics labeling unreliable. FreeAct addresses this challenge by learning a discrete latent action space from large-scale multi-lab robot videos and aligning it with camera-frame end-effector SE(3) motion and gripper-state changes. The resulting latent action tokenizer captures task-relevant motion while reducing reliance on embodiment-specific visual artifacts, enabling generated target-domain videos to be pseudo-labeled with reliable latent action tokens. These tokens are first injected into VLA pretraining and then used, together with source-domain action supervision, to adapt the policy to unseen environments. Empirically, FreeAct improves success from 2\% to 57\% under target-domain shift using generated videos alone, and further to 74\% through MLLM-guided self-evolution on the policy's own rollouts, closing the adaptation loop without expert demonstrations. Further analysis shows that the learned latent action space forms a kinematically structured and transferable representation, establishing generated videos as a scalable source of supervision for robot manipulation.

Tools from random matrix theory have become central to deep learning theory, using spectral information to provide mechanisms for modeling generalization, robustness, scaling, and failure modes. While often capable of modeling empirical behavior, practical computations are limited by matrix size, often imposing a restriction to models that are too small to be realistic. This motivates the inference of properties of larger models from the behavior of smaller ones. Free decompression (FD) is a recently proposed method for extrapolating spectral information across matrix sizes, but its utility is currently limited by strong assumptions that preclude its implementation on more realistic machine learning (ML) models. We use algebraic spectral curve theory to provide a general FD methodology for spectral densities whose Stieltjes transform satisfies an algebraic relation, a modeling assumption that is more likely to hold in practice. This recasts FD as an evolution along spectral curves which can be readily integrated. Our framework enables the expansion of spectral densities that have multiple or multi-modal bulks, that exist at multiple scales, and that contain atoms, all characteristic of real-world data and popular ML models. We demonstrate the efficacy of our framework on models of interest in modern ML, including Hessian and activation matrices associated with neural networks and large-scale diffusion models.

Hamiltonian Neural Networks and related structure-preserving dynamics models encode conservation laws, but their scalar Hamiltonian parameterizations inherit the spectral bias of deep networks, limiting their ability to learn stiff multi-timescale systems with coupled fast and slow dynamics. We introduce the Frequency-Structured Hamiltonian Neural Network (FS-HNN), which decomposes the system Hamiltonian into learned components trained on frequency-filtered views of observed trajectories, then recombines them into a single scalar Hamiltonian so that the learned ODE dynamics remain Hamiltonian by construction. For PDEs with unknown or problem-dependent structure, FS-HNN represents Hamiltonians as neural functionals, learns the action of the dynamics operator, and uses projection to enforce conservative structure when appropriate. Across ODE and PDE benchmarks, including the Fermi--Pasta--Ulam--Tsingou chain, shallow water equations, and incompressible Taylor--Green vortex, FS-HNN achieves improved long-horizon rollout accuracy and more accurate energy behavior than existing structure-preserving baselines.

Text-to-video generation has made remarkable progress with diffusion models and transformer-based video architectures, yet generating coherent long videos from evolving textual descriptions remains challenging. Training long-video models from scratch requires substantial computation and large-scale multi-event data, while training-free multi-prompt generation must activate new prompt semantics without breaking temporal continuity. We identify a frequency-agnostic boundary coupling problem in existing transition strategies: weak coupling causes abrupt changes, flicker, and background jumps, whereas overly strong or coarse coupling may over-preserve source semantics and produce ghosting or duplicated subjects around prompt boundaries. To address this problem, we propose FreqSync, a training-free framework that formulates prompt transitions as frequency-selective boundary coupling. FreqSync combines Frequency-Synchronized Source-Conditioned Attention (FS-SCA) with Residual-Clipped Spectral Stitching (RCSS), enabling adjacent segments to share stable low- and mid-frequency structure and motion cues while keeping high-frequency prompt-specific details target-dominant under explicit residual constraints. Extensive experiments show that FreqSync achieves state-of-the-art transition smoothness and prompt alignment while maintaining strong perceptual video quality, demonstrating its effectiveness for coherent multi-prompt long video generation.

Large language models are increasingly used to estimate uncertain quantities from context, raising the question of whether their probabilistic outputs are internally consistent. A basic test of self-consistency is whether these estimates adhere to the law of total probability. We investigate this requirement through the lens of binary conditioning trees, which recursively partition a population into increasingly fine-grained subsets and expose multiple ways of estimating the same aggregate quantity. This construction yields a basic self-consistency check: marginal estimates must agree with prior-weighted aggregations of conditional estimates over any partition of the population. We find that current models systematically violate this requirement. In a case study on persona prompting, prior-weighted aggregates are consistently better aligned with human population statistics than direct estimates. Notably, this benefit of specificity persists even when the prior weights are themselves estimated by the LLM. Turning this discrepancy into a general evaluation criterion, we propose a family of self-consistency checks for LLMs grounded in the law of total probability. By evaluating these checks across frontier models, we show that failures of statistical consistency are widespread and not confined to persona prompting. Together with a benchmark dataset, our work provides a testbed for understanding when in-context learning can be treated as conditional inference.

While it is generally understood that zeroth-order (ZO) algorithms have an extra dependency on their number of iterations for any choice of parameters, compared to their first-order (FO) counterparts, in this work, we show that under several conditions, in expectation, ZO methods do not suffer from extra dimension dependencies in their convergence rates with respect to their FO counterparts. We look at optimisation algorithms from the dynamical systems perspective and analyse the conditions under which one can formulate the average of a ZO algorithm as the average of its FO counterpart with bounded perturbations with values dependent on design parameters. Then, using input-to-state stability properties, we show ZO methods follow the same decay rate as their FO counterparts and converge to a neighbourhood of the fixed point of FO methods, where its radius depends on the bound of the norm of the perturbations, which can be made arbitrarily small. The theoretical findings are illustrated via numerical examples.

Discrete diffusion models have emerged as a promising direction for vision-language tasks, offering bidirectional context modeling and theoretical parallelization. However, their practical application is severely hindered by a train-inference discrepancy, which leads to catastrophic error cascades: initial token errors during parallel decoding pollute the generation context, triggering a chain reaction of compounding errors and leading to syntactic errors and semantic hallucinations. To address this fundamental challenge, we reframe the generation process from passive denoising to active refining. We introduce ReDiff, a refining-enhanced diffusion framework that teaches the model to identify and correct its own errors. Our approach features a two-stage training process: first, we instill a foundational revision capability by training the model to revise synthetic errors; second, we implement a novel online self-correction loop where the model is explicitly trained to revise its own flawed drafts by learning from an expert's corrections. This mistake-driven learning endows the model with the crucial ability to revisit and refine its already generated output, effectively breaking the error cascade. Extensive experiments demonstrate that ReDiff significantly improves the coherence and factual accuracy of generated content, enabling stable and efficient parallel generation far superior to traditional denoising methods.

Industry-scale video and live-streaming moderation imposes requirements that are difficult to satisfy with generic pretrained public models or external APIs, including adaptation to platform-specific data distributions, policy-specific objectives, and product-level safety constraints. As a result, platforms must undertake internal model development, naturally turning to shared public research for guidance. However, existing multimodal foundation-model studies primarily report architectures, training recipes, data scaling strategies, and benchmark results, but provide less systematic guidance on how failures should be localized and translated into targeted model-development interventions. Interventions are essential because deployment failures are rarely self-explanatory. Similar failures can originate from different causes. Without targeted interventions, improvement reduces to heuristic trial-and-error, where benchmark improvements are weakly attributable, and failures are difficult to trace to their underlying causes. To address this gap, we present a diagnostic methodology for industry-scale Audio-Visual-Language Models~(AVLM) development. The methodology maps model failures into a taxonomy of observable failure signatures and links each class of failure to an intervention space. We instantiate this methodology across the development and alignment lifecycle of an AVLM foundation model for a large-scale video and live-streaming platform. The resulting system supports over 100 regions and is designed for noisy, ambiguous, and highly diverse content drawn from global platform traffic.

Online Continual Learning (OCL) requires models to learn from non-stationary data streams under a strict single-pass constraint, making them highly susceptible to catastrophic forgetting. While existing studies have explored various strategies from data, architecture, and optimization perspectives, the optimizer often directly uses generic approaches such as SGD and AdamW. In this work, we reveal a new link between optimization dynamics and OCL needs by recasting Polyak-Ruppert averaging as the engine of "plasticity" and Primal averaging as the anchor of "stability". This inspires our novel optimizer, SPIN (Stability-Plasticity INterpolation), which explicitly decouples plasticity and stability by interpolating between a fast-moving "plasticity" sequence and an adaptive "stability" sequence. We theoretically demonstrate that SPIN implicitly employs an inverse Hessian approximation, providing crucial Tikhonov regularization to damp noisy, single-sample Hessian estimates and mitigate forgetting. More importantly, SPIN can be seamlessly integrated existing OCL methods by taking resulted gradients as input and replacing standard optimizers. Extensive experiments show that using SPIN instead of standard optimizers yields consistent and significant performance gains across multiple OCL approaches, including the challenging rehearsal-free setting and under strict memory constraints.

Generating humorous memes is a challenging multimodal task that moves beyond direct image-to-caption supervision. It requires a nuanced reasoning over visual content, contextual cues, and subjective humor. To bridge this gap between visual perception and humorous punchline creation, we propose HUMOR, a novel framework that guides VLMs through hierarchical reasoning and aligns them with group-wise human preferences. First, HUMOR employs a hierarchical, multi-path Chain-of-Thought (CoT): the model begins by identifying a template-level intent, then explores diverse reasoning paths under different contexts, and finally anchors onto a high-quality, context-specific path. This CoT supervision, which traces back from ground-truth captions, enhances reasoning diversity. We further analyze that this multi-path exploration with anchoring maintains a high expected humor quality, under the practical condition that high-quality paths retain significant probability mass. Second, to capture subjective humor, we train a pairwise reward model that operates within groups of memes sharing the same template. Following established theory, this approach ensures a consistent and robust proxy for human preference, even with subjective and noisy labels. The reward model then enables a group-wise RL optimization, providing a theoretical guarantee for monotonic improvement within the trust region. Extensive experiments show that HUMOR empowers various VLMs with superior reasoning diversity, more reliable preference alignment, and higher overall meme quality. Beyond memes, our work presents a general paradigm for open-ended, human-aligned multimodal generation, where success is guided by comparative judgment within coherent output groups.

Time series foundation models (TSFMs) have achieved notable progress in classification tasks through large-scale pretraining. However, existing methods rely on implicit parameter fine-tuning, where task-relevant information is entangled with abundant irrelevant knowledge in vast parameter spaces. This may hinder TSFMs from precisely activating useful knowledge for downstream classification. To this end, we propose REKIN, a plug-and-play knowledge-enhanced framework for time series classification built upon pretrained TSFMs within a retrieval-to-recognition paradigm. Unlike traditional parameter-only approaches, REKIN introduces a decoupled query encoder to extract discriminative patterns and retrieve instance-level numerical knowledge from a constructed knowledge base. The retrieved explicit knowledge is then fused with the TSFM’s implicit parametric embeddings, enabling joint knowledge transfer at both the parameter and feature levels. In this way, REKIN allows TSFMs to recall task-relevant patterns learned during pretraining, improving the efficiency of knowledge utilization in downstream transfer. Experiments on 158 time series datasets show that REKIN consistently improves the classification performance of state-of-the-art TSFMs. Ablation and knowledge-enhanced studies further validate the effectiveness of the proposed framework.

Engineering value is often created not by producing a single correct answer, but by improving a feasible artifact under hard constraints. We introduce **Frontier-Eng**, a benchmark for generative optimization agents that repeatedly edit executable engineering artifacts, receive frozen verifier feedback, and maximize the best feasible score found within a fixed budget. Frontier-Eng contains $47$ tasks across five engineering categories, spanning kernels, cryptography, quantum circuits, inventory and scheduling, robotics and control, optics and communication systems, structural design, reactions, and sustainable infrastructure. Each task provides a feasible starting solution, an editable artifact, and a read-only evaluator with verifier-parsed scoring, enabling cross-domain comparison without reducing engineering quality to binary pass/fail completion. We evaluate frontier models under OpenEvolve and compare search frameworks across the full task set. Current agents exhibit clear iterative improvement: **gpt-5.4** obtains the best average rank under OpenEvolve. Long-horizon runs reveal a dual **inverse-law pattern**: improvement events become rarer with iteration, and improvement magnitudes shrink with improvement count. At the same time, improvement is uneven across domains and remains sensitive to search depth, exposing a concrete frontier for engineering-oriented agents under realistic constraints and verifier-grounded feedback.

In this work, we study nonconvex–strongly convex online bilevel optimization (OBO) using only first-order oracle. Existing OBO algorithms are mainly based on hypergradient descent, which requires access to a Hessian-vector product (HVP) oracle and potentially incurs high computational costs. By reformulating the original OBO problem as a single-level online problem with inequality constraints and constructing a sequence of Lagrangian function, we eliminate the need for HVPs arising from implicit differentiation. Specifically, we propose a fully first-order algorithm for OBO, and provide theoretical guarantees showing that it achieves regret of $O(1 + V_T + H_{2,T})$ with a total of $O(T\log T)$ iterations, where $V_T$ measures the variation in function values and $H_{2,T}$ characterizes the drift variation of the inner-level optimal solution. We also establish a sublinear regret bound under the single-loop structure by introducing additional gradient-variation terms. Furthermore, we develop an improved variant with an adaptive inner-iteration scheme, which removes the dependence on $H_{2,T}$ and achieves regret of $O(\log T + V_T)$. Finally, under the stochastic OBO setting, we establish the regret bound for the fully first-order algorithm, i.e., $O(T^{2/3}(1 + \sigma^2) + V_T + H_{2,T})$. Numerical experiments demonstrate the feasibility of our algorithm and support our theoretical findings.

As vision-language models (VLMs) transition from frozen-encoder pipelines to native multimodal architectures, the visual channel increasingly serves as both a perceptual input and a carrier of rendered text that competes with user prompts. Current single-attack, pixel-centric evaluations fail to capture this complexity: they collapse distinct visual attack surfaces into a single scalar and ignore how architectural shifts alter vulnerability. We introduce \textbf{PaRMA}, an evidence-based auditing framework that defines pathway-conditioned risk over two primary image-level interventions: pixel perturbation and rendered-text insertion. Instead of arbitrary attack enumeration, PaRMA uses targeted diagnostics to expose architecture-specific blind spots. Notably, we uncover a \textit{see-but-not-deceived} signature where native VLMs attend heavily to typographic overlays but visually reject them, and we identify severe gradient dispersion in deep-fusion models. To exploit these findings, we instantiate \textbf{FusionAudit}, introducing two targeted attacks: \textbf{MANTIS} (margin-aware pixel perturbation) and \textbf{Nat-Evid} (host-congruent semantic insertion). Evaluating 18 open-weight VLMs across a novel nativeness taxonomy, we demonstrate that legacy single-attack rankings (e.g., PGD) are highly unstable for native models. Our diagnostic-driven attacks substantially raise the estimated residual risk, proving that PaRMA provides a falsifiable, evidence-driven auditing methodology to uncover vulnerabilities systematically hidden by standard benchmarks.

Modern LLMs/VLMs have largely converged to a unified architecture: representing heterogeneous inputs as a single token sequence and processing it with a deep stack of generic operators such as attention and feed-forward layers. This sequence-first interface simplifies system designs and thus enables extensive algorithmic and hardware optimization, making it a compelling blueprint for modern model development. In contrast, 3D multi-modal fusion models still rely on complex and specialized designs such as dense feature representations, sparse 3D convolutions, deformable attention, view transformations, etc. In this paper, we ask: \textbf{Can 3D multi-modal fusion embrace the same design path as modern LLMs/VLMs and benefit from their mature and rapidly evolving ecosystem?} We answer this affirmatively and propose \textbf{FusionNeXt}, a modern multi-modal fusion paradigm that (i) unifies camera and LiDAR features into a shared token representation, (ii) serializes tokens into locality-preserving 1D sequences, and (iii) performs feature fusion using a deep stack of vanilla LLM/VLM-style blocks---FlashAttention, pre-norm residuals, SwiGLU FFNs, etc. The proposed paradigm enables standard sequence modeling tools to be applied directly to 3D multi-modal fusion, resulting in fast inference, strong performance, and generalization across tasks. FusionNeXt achieves state-of-the-art (SOTA) results on the nuScenes 3D detection and Occ3D occupancy benchmarks while delivering high inference throughput, suggesting a scalable direction for 3D perception. We will open-source our code.

Competitive multi-agent reinforcement learning in imperfect-information games requires agents to act under partial observability and against adversarial opponents, necessitating stochastic policies. While self-play reinforcement learning with Proximal Policy Optimization (PPO) has achieved strong empirical success, its standard advantage estimator, generalized advantage estimation, suffers from additional variance due to the sampling of stochastic future actions. This variance is amplified in equilibrium self-play because of the stochastic nature of the equilibrium policy and persists even when the critic is exact. We address this bottleneck by introducing $Q$-boosting, a variance-reduced advantage estimator based on a centralized action-value critic, and propose Variance-Reduced Policy Optimization (VRPO), incorporating this new estimator. The algorithm replaces sampled multi-step backups with a multi-step Expected SARSA$(\lambda)$ trace, computing policy expectations at each step to average out action-sampling noise, while retaining PPO's clipped objective and on-policy actor updates. Empirically, VRPO consistently achieves strong performance from mid-sized to large-scale games including Dou Dizhu and Heads-Up No-Limit Texas Hold’em.

Reconstruction error is the standard signal for unsupervised time-series anomaly detection and has been widely adopted by time-series foundation models (TFMs). It is, however, vulnerable to input noise: the resulting score tracks the noise level itself rather than whether a sample is truly anomalous, and this failure is especially pronounced in domains where the noise envelope itself carries information about the underlying process state. We propose the *adaptation gap* $\text{Diff} = L_{FT} - L_{ZS}$, the difference between fine-tuned and zero-shot reconstruction losses, as a noise-robust complement. We prove that its conditional expectation cancels the input-noise variance and yields a calibration-anchored, population-level mean-separation guarantee. Building on these results we introduce **GAPS** (Gradient-Aware Adaptation-Gap Scoring), which selectively augments $L_{FT}$ with $\text{Diff}$ under a calibration-only routing layer with no label-tuned hyperparameters. Across a synthetic suite, an HRV benchmark whose normal class is intrinsically noisier than its anomaly class, and TSB-AD-U with a pre-registered noise-stress condition, GAPS preserves performance under standard conditions and yields substantial gains under noise stress.

Born-Oppenheimer molecular dynamics delivers quantum accurate trajectories by evolving nuclei on the electronic ground-state surface while retaining electronic observables such as dipoles and polarizabilities. In Kohn-Sham density functional theory, this typically relies on an iterative self consistent field solver at each time step. Machine learning can predict electronic quantities or improve SCF initialization, yet dynamics remain sensitive to residual self consistency error, which can yield non conservative forces and energy drift, and repeated diagonalization limits differentiable and accelerator efficient implementations. We propose a gauge-symmetric dual Lagrangian framework that avoids SCF loops by propagating the physically meaningful electronic state as the occupied subspace and its density projector on a Grassmann manifold. A residual driven update stabilized by Rayleigh type dissipation with time decaying inertia yields a closed evolution of the subspace, its conjugate momentum, and the projector. Coupled with a neural gauge-symmetric Hamiltonian with analytic coordinate derivatives, the method provides closed form forces with Pulay corrections and enables stable quantum BOMD on molecular simulation benchmarks.

This paper proposes a novel crowd counting approach, the Gaussian Density Splatting Network (GDSNet). Unlike methods that rely on conventional, grid-based density maps and are sensitive to spatial resolution, GDSNet represents a crowd as a superposition of continuous 2D Gaussian primitives. Our approach is built upon two key contributions. First, we introduce a control-point-based fitting mechanism to structure the prediction of the Gaussian parameters. We design a method to allocate a set of control points that define local regions, from which features are pooled to regress each primitive's parameters. Second, we adapt a differentiable Gaussian Splatting framework to the counting task by parameterizing each primitive with geometric parameters and a scalar density mass. This formulation allows the network to be trained end-to-end via spatial matching of differentiably rendered density maps, naturally providing both local density supervision and global count optimization. Extensive evaluations on four standard benchmarks show GDSNet consistently outperforms the state of the art. Code will be released upon acceptance.

Immersive video communication requires photorealistic, render-efficient, and compact dynamic scene representations. 3D Gaussian Splatting (3DGS) offers a promising representation, but dynamic 3DGS remains difficult to compress due to dense primitives and spatiotemporal redundancy. Anchor-based formulations improve compactness with sparse scaffolds that share geometry and appearance across primitives. However, existing designs often rely on deforming a single canonical scaffold and condition each primitive on its associated anchor in isolation, limiting their ability to handle non-local dynamics and disocclusion while under-exploiting inter-anchor correlations, particularly in motion- or texture-dense regions. To address these limitations, we propose SAGA, a volumetric video codec built upon Sparse Anchor-assisted GAussian splatting representations. SAGA represents dynamic 3D scenes using hierarchically organized sparse 4D anchors, where coordinate-based INR decoders generate fine anchors and Gaussian primitives from inter-anchor interpolations, enabling compact parameter sharing across spatiotemporal structures. For long-range dependencies among unstructured anchors, we further introduce fixed-size memory slots with orthogonality-informed updates for accurate entropy-context modeling. Experiments show SAGA outperforms the state-of-the-art 3DGS-based codec GIFStream, achieving BD-rate reductions of 84.54% on Neu3D, 76.71% on Panoptic Sports, and 83.94% on MPEG MIV.

Training modern neural networks often relies on large learning rates, operating at the edge of stability, where the optimization dynamics exhibit oscillatory and chaotic behaviour. Empirically, this regime often yields improved generalization performance, yet the underlying mechanism remains poorly understood. In this work, we represent stochastic optimizers as random dynamical systems, which often converge to a fractal attractor set (rather than a point) with a smaller intrinsic dimension. Building on this connection and inspired by Lyapunov dimension theory, we introduce a novel notion of dimension, coined the `sharpness dimension', and prove a generalization bound based on this dimension. Our results show that generalization in the chaotic regime depends on the complete Hessian spectrum and the structure of its partial determinants, highlighting a complexity that cannot be captured by the trace or spectral norm considered in prior work. Experiments across various MLPs and transformers validate our theory while also providing new insights into the recently observed phenomenon of grokking.

Empirical evidence suggests that large neural networks rarely make effective use of all available parameters, with learned solutions exhibiting pronounced sparsity. Despite this ubiquity, existing generalization theory only partially explains how such sparsity influences statistical performance. In this paper, we derive non-asymptotic excess risk bounds for deep neural networks whose layer-wise weight matrices are restricted by $\ell_p$ quasi-norm ($0

We study high-dimensional observations $Z_i \in \mathbb{R}^p$ satisfying that $E[Z_i] = X(t_i)$, where $X(t)$ parameterizes a manifold on $0 \le t < 2\pi$, with the goal of recovering the latent phases $t_i$; a motivating example is estimating object-viewing angles from photographs. Spectral seriation is a standard phase-recovery method: given a similarity matrix $S$ with degree matrix $D$, it uses the leading eigenvectors of the normalized Laplacian $D^{-1/2} S D^{-1/2}$, whose continuum limit is linked to the manifold Laplacian, to estimate $t_i$. Recent applications show that generalized Laplacians involving $D^{-\alpha}$ can outperform the normalized Laplacian on graphs with heterogeneous degrees, raising the central question of whether the same advantage holds for spectral seriation. We address such question by introducing and comparing three principled generalizations drawn from manifold learning and spectral seriation: a shifted generalized Laplacian, a graph-difference Laplacian, and an extra-normalized generalized Laplacian. The extra-normalized generalized Laplacian is the most robust to $\alpha$ and achieves the highest recovery accuracy, identifying extra normalization as the key stabilizing step. We establish a consistency result unveiling that eigenspace perturbation controls the estimation error up to rotation and reflection, giving a theoretical basis for the method. Experiments on noisy closed curves confirm more accurate phase recovery under a suggested choice of $\alpha$; on image data, the method recovers photo angles with the best accuracy. Our results demonstrate that extra normalization is a key component for stable generalized spectral seriation in closed-curve recovery problems.

Shapley value and its priority-aware extensions are widely used for valuation in machine learning, but existing methods require pairwise priority to be binary and acyclic, a restriction spectacularly violated in real-data examples such as aggregated human preferences and multi-criterion comparisons. We introduce the generalized priority-aware Shapley value (GPASV), a random order value defined on arbitrary directed weighted priority graphs, in which pairwise edges penalize rather than forbid order violations. GPASV covers a range of classical models as boundary cases. We establish GPASV through an axiomatic characterization, develop the associated computational methods, and introduce a priority sweeping diagnostic extending PASV's. We apply GPASV to LLM ensemble valuation on the cyclic Chatbot Arena preference graph, illustrating that priority-aware valuation is not a one-button operation: different balances of pairwise graph priority versus individual soft priority produce substantively different valuations of the same data.

Individual-level wildlife identification often suffers from data scarcity, as varying observations of the same animal under diverse poses, viewpoints, and motions are rarely available. Image-to-video (I2V) generation offers a promising way to mitigate this limitation by synthesizing additional observations from a single reference image. However, existing I2V models mainly emphasize global layout, semantics, and motion, and therefore often fail to preserve fine-grained local appearance cues that distinguish one wildlife individual from another, such as fur texture, stripe boundaries, spot configurations, and contour transitions. We observe that these identity-critical cues are closely related to high-frequency information, which is essential for distinguishing individuals. To address this challenge, we propose WildIcon, a high-frequency-guided I2V framework for wildlife individual consistency. Specifically, WildIcon introduces a frequency-aware identity encoding branch that extracts individual-specific high-frequency cues from the reference image. Combined with isolated foreground information, the resulting identity tokens are then injected into cross-attention blocks as identity conditioning. Building on a frozen backbone with lightweight identity adaptation, WildIcon strengthens fine-grained identity preservation while retaining the motion controllability and semantic fidelity of the base I2V model. In addition, to support the training and evaluation of wildlife individual-consistent I2V, we construct WildlifeVid, a wildlife-centric video dataset with high-quality, temporally coherent clips and individual-level identity labels. Experiments on both I2V and downstream animal re-identification (ReID) show that WildIcon achieves stronger individual consistency than existing baselines and provides useful, reliable synthetic data to improve downstream ReID performance.

Unlike real-world datasets, utilizing synthetic data from generative models as train dataset necessitates a rigorous consideration of the training data's informativeness. In this paper, we propose a novel framework designed to steer diffusion models based on Bayesian epistemic uncertainty to train the target model. To ensure computational tractability, we estimate the Bayesian Active Learning by Disagreement (BALD) score by applying the Laplace approximation specifically to the last-layer parameters of the downstream model. Furthermore, we incorporate Feynman-Kac Steering (FKS) for diffusion to enhance the structural diversity of the synthetic samples and mitigate the emergence of undesirable artifacts. Extensive experiments on Imagenette and ImageNet-100 for the classification task and MS-COCO for the object detection task demonstrate that our method significantly outperforms existing baselines in downstream accuracy and OOD generalization. Furthermore, our analysis of pairwise distances between inter-class and intra-class confirms that our method effectively expands the downstream model's knowledge manifold by generating diverse and informative training signals.

Autoregressive video diffusion models have demonstrated effectiveness in interactive game generation, with Minecraft gameplay serving as a representative application. To faithfully simulate gameplay, a model must generate natural content when exploring new scenes while preserving spatial consistency when revisiting explored areas. Under limited computation budgets, it must compress and exploit historical cues within a finite context window, which exposes a trade-off: models relying solely on temporal context offer flexible exploration but suffer from poor revisit consistency, whereas adding spatial memory strengthens consistency but may degrade new scene generation quality when the model over-relies on sparse or unreliable spatial context. We present GeoMemory, a learning framework that pairs training protocols with a geometry-indexed spatial memory. Specifically, our Hybrid Training exposes the model to both exploration and revisitation regimes, guiding the model to rely on temporal memory in new scenes while effectively incorporating spatial memory upon revisits. Chained Forward Training creates larger pose variations and encourages reliance on spatial memory for maintaining consistency. For spatial memory, we integrate Point-to-Frame Retrieval with an incremental 3D cache generated by VGGT, enabling constant-time retrieval of relevant historical context regardless of sequence length. Extensive experiments demonstrate that GeoMemory achieves superior performance in both long-term spatial consistency and visual quality in new scenes with real-time interaction.

Large language models often produce fluent reasoning chains that arrive at incorrect answers, and these failures are hard to detect from output probabilities alone. We show that a strong signal of correctness is reflected in the geometry of internal computation: hidden-state trajectories across transformer layers differ between correct and incorrect reasoning, with correct solutions following smooth, efficient paths and failures exhibiting elevated curvature, abrupt directional changes, and increased geodesic deviation. We capture this signal through VANE, five parameter-free geometric features (velocity, acceleration/jerk, curvature, geodesic deviation, and token coherence) computed from a single forward pass over layer-wise hidden states. Across six models (1.5B to 72B parameters) and five benchmarks spanning mathematics, code, and verbal reasoning, VANE achieves 70.7 to 96.6 AUROC and exceeds token log-probability on all 30 model–benchmark pairs (mean +25 percentage points, up to +50 percentage points). The geometric signal is near-independent of output confidence (r = -0.26), and a base-model control shows it emerges with instruction tuning. As a downstream application, filtering geometrically unstable outputs raises accuracy to 96.1% at 50% coverage on a 4-bit quantized 72B model. These results show trajectory geometry is a reliable single-pass indicator of reasoning correctness, even when the model is confidently wrong.

Generalizable 3D Gaussian Splatting aims to predict renderable Gaussian scenes from sparse images without per-scene optimization at test time. Existing feed-forward methods improve this problem mainly by strengthening the evidence given to the predictor, such as multiview aggregation, geometric cues, auxiliary supervision, and surface priors. Yet the prediction itself is usually learned as a direct map to the final Gaussian scene. This endpoint-centered objective compresses scene layout, local geometry, visibility, opacity, orientation, and appearance into a single end-point target, leaving the construction path of the Gaussian scene implicit. We propose GRiF, a geometry-aware flow matching framework that augments endpoint target supervision with conditional transport in scene space. Starting from a controlled source scene, GRiF learns a view-conditioned velocity field that progressively evolves a hierarchical Gaussian state toward a renderable target. The hierarchy organizes updates from coarse layout to fine attributes, while geometry-aware paths respect the native structure of Gaussian variables with rotation-aware paths for orientation variables. Experiments on RealEstate10K and ACID, with zero-shot transfer evaluated on ScanNet, DL3DV, and DTU, show improved in-domain reconstruction and stronger cross-domain performance.

The explosive demand for interactive Large Language Model serving has highlighted the management of the Key-Value cache's dynamic memory footprint as a critical area for performance optimization in inference engines. Modern inference systems overwhelmingly rely on time-centric scheduling heuristics, such as Shortest Job First. However, their theoretical optimality is rooted in traditional schedule modeling, failing to capture the highly dynamic, 2D spatio-temporal geometric growth specific to LLM inference mechanisms. To resolve this, we propose the geometry-aware online scheduling by introducing the Smallest Volume First (SVF) algorithm and its highly efficient variant, 1-bit SVF. Theoretically, we provide a rigorous mathematical foundation for our approach. Utilizing a novel proof methodology, we tighten the worst-case competitive ratio ($\text{CR} \le 48 \rightarrow \text{CR} \le 5$) for SVF with known output lengths. Building upon this core breakthrough, we complete a comprehensive theoretical taxonomy analyzing our algorithms across different traffic scenarios and information availability. Practically, we seamlessly integrate our approach as a plug-and-play layer in vLLM. Extensive evaluations on Llama-3.1 models demonstrate comprehensive performance gains: SVF delivers strong reductions in both average and tail latency, while 1-bit SVF, with merely a single bit information, achieves competitive throughput and latency. This work establishes a theoretically sound and empirically proven approach for resolving memory-constrained scheduling in modern LLM deployments. To facilitate future research, our code is available at \url{https://anonymous.4open.science/r/LLM-inference-scheduling-7970}.

Similarity measures such as canonical correlation analysis (CCA), representational similarity analysis (RSA) and center kernel alignment (CKA), are widely used to compare the representational geometries used by different neural networks to solve the same task. Yet, because existing methods compare the extrinsic geometry of neural representations in state space, rather than their intrinsic geometry, they may fail to capture subtle yet crucial distinctions between fundamentally different computations performed by biological and artificial neural networks. Here, we introduce metric similarity analysis (MSA), a novel method which leverages tools from Riemannian geometry to compare the intrinsic geometry of neural manifolds. Within our mathematical framework, we derive several properties of MSA, including coordinate, scale and rotation invariances. We show that MSA can be used to i) disentangle neural computations of deep networks in rich vs. lazy learning regimes, ii) compare the computation-through-dynamics of recurrent neural networks and state-space models during neuroscience working memory tasks, and iii) investigate the statistical manifolds of diffusion models. Overall, we introduce a mathematically grounded and broadly applicable framework to understand the mechanisms behind neural computations by contrasting their Riemannian geometries, with broad applicability to both neuroscience and machine learning.

When language models lack relevant knowledge for a given query, they frequently generate plausible responses that can be hallucinations, rather than admitting being agnostic about the answer. Retraining models to reward admitting ignorance can lead to overly conservative behaviors and poor generalization due to scarce evaluation benchmarks. We propose a post-hoc framework, Conformal Abstention (CA), adapted from conformal prediction (CP) to determine whether to abstain from answering a query. CA provides finite-sample guarantees on both the probability of participation (i.e., not abstaining) and the probability that the generated response is correct. Importantly, the abstention decision relies on prediction confidence rather than the non-conformity scores used in CP, which are intractable for open-ended generation. To better align prediction confidence with the model's ignorance, we introduce a calibration strategy using representation geometry within the model to measure knowledge involvement in shaping the response. Experiments demonstrate that we improve selective answering significantly with 75% conditional correctness.

Continual post-training aims to extend large language models (LLMs) with new knowledge, skills, and behaviors, yet it remains unclear when sequential updates enable capability transfer and when they cause catastrophic forgetting. Existing methods mitigate forgetting through sequential fine-tuning, replay, regularization, or model merging, but offer limited criteria for determining when incorporating new updates is beneficial or harmful. In this work, we study LLM continual post-training through three questions: What drives forgetting? When do sequentially acquired capabilities transfer or interfere? How can compatibility be used to control update integration? We address these questions through task geometry: we represent each post-training task by its parameter update and study the covariance geometry induced by the update. Our central finding is that: forgetting can be considered as a state-relative update-integration failure, it arises when the covariance geometries induced by tasks misalign with the geometry of the evolving model state. Sequential updates transfer when they remain compatible with the model state shaped by previous updates, and interfere when state-relative geometry conflict becomes high. Motivated by this finding, we propose Geometry-Conflict Wasserstein Merging (GCWM), a data-free update-integration method that constructs a shared Wasserstein metric via Gaussian Wasserstein barycenters and uses geometry conflict to gate geometry-aware correction. Across Qwen3 0.6B--14B on domain-continual and capability-continual settings, GCWM consistently outperforms data-free baselines, improving retention and final performance without replay data. These results identify geometry conflict as both an explanatory signal for forgetting and a practical control signal for LLM continual post-training.

Multimodal molecular representation learning aligns heterogeneous chemical and cellular signals to predict molecular properties. The dominant paradigm is instance-level contrastive alignment, such as InfoNCE and VICReg, which implicitly treats every modality as a point cloud in a shared Euclidean space. This paradigm overlooks two structural properties of molecular-cellular data. \textit{First}, each modality, from atomic fingerprints to cell morphology to transcriptomic response, carries its own intrinsic geometry; directly pulling individual embeddings together can distort the relational structure each modality has learned to express. \textit{Second}, the modalities form a directed biological cascade, from molecular structure to conformation, gene perturbation, cellular phenotype, and transcriptomic response, whose branching grows exponentially with depth, a regime that Euclidean spaces cannot embed without distortion. We therefore propose \textbf{GeoPMR}, a pre-training framework that addresses both properties through two complementary geometric modules: Metric-Guided Gromov-Wasserstein alignment (MGW) and a Hyperbolic Hierarchical Module (HHM). \textbf{MGW} equips each modality with a learnable diagonal Riemannian metric and aligns distance matrices rather than individual embeddings via entropic optimal transport, so cross-modal alignment preserves relational rather than merely pointwise structure. \textbf{HHM} lifts all modality latents onto the Poincaré ball and applies a probabilistic parent-child and sibling contrastive loss along the cascade, exploiting the exponential volume growth of negatively curved space to embed the hierarchy with low distortion. On seven molecular property prediction benchmarks spanning toxicity, bioactivity, and ADME, GeoPMR sets a new state of the art by improving the classification AUC by 1.4\% and regression MAE by 1.7\%.

Worldwide image geolocalization aims to predict precise GPS coordinates for images captured anywhere on Earth, which is challenging due to the large visual and geographic diversity. Recent methods mainly follow two paradigms: retrieval-based approaches that match queries against a reference database, and generation-based approaches that directly predict coordinates using Large Vision-Language Models (LVLMs). However, we observe distinct error profiles between them: retrieval excels at fine-grained instance matching, while generation offers robust semantic reasoning. This complementary heterogeneity suggests that no single paradigm is universally superior. To harness this potential, we propose GeoRouter, a dynamic routing framework that adaptively assigns each query to the optimal paradigm. GeoRouter leverages an LVLM backbone to analyze visual content and provide routing decisions. To optimize GeoRouter, we introduce a distance-aware preference objective that converts the distance gap between paradigms into a continuous supervision signal, explicitly reflecting relative performance differences. Furthermore, we construct GeoRouting, the first large-scale dataset tailored for training routing policies with independent paradigm predictions. Extensive experiments on IM2GPS3K and YFCC4K demonstrate that GeoRouter significantly outperforms state-of-the-art baselines. Our dataset, code, and checkpoints are publicly available to facilitate future research.

Vision-Language-Action (VLA) models often struggle with precise manipulation due to their lack of explicit spatial awareness. To bridge this gap, we propose GIVLA, a framework that internalizes geometric priors into the VLA backbone through a deep-coupled architecture. GIVLA implements a two-stage gradient-informed training paradigm to resolve task-level interference and ensure precise execution, a design grounded in the perception-action duality identified through our analysis of training dynamics. Extensive experiments on LIBERO and physical robot platforms demonstrate that GIVLA achieves superior accuracy and robustness with high parameter efficiency.

While talking head generation has advanced rapidly, generating natural listener behavior in dyadic conversations, which know when to react, how to react, and with what type of response, remains underexplored. Existing dyadic datasets lack fine-grained listener reaction annotations, and prevailing evaluation metrics inherited from talking-head and video generation measure visual realism rather than whether a listener reacted appropriately. We address these gaps along three aspects. First, we curate a listening-head-specific dataset built from RealTalk and Seamless Interaction, comprising approximately 147 hours of paired speaker–listener videos with 64,557 event-level reaction annotations across six categories: nodding, head shaking, smiling, laughing, frowning, and surprised. Second, we introduce an audio-driven baseline built on a flow-matching transformer, namely GLARE, with prosody conditioning derived from Qwen2-Audio and a temporal reaction loss that explicitly supervises frame-wise reactions. Third, we propose a reaction-oriented evaluation protocol that jointly measures reaction occurrence (R-F1), temporal alignment (R-tIoU), asymmetric temporal deviation (R-ATD), and reaction-region visual quality (R-FID), giving a more behaviorally grounded assessment than visual-quality-only metrics. Experiment results show consistent gains over prior listening-head methods in both visual fidelity and reaction-level metrics, suggesting that reaction-aware data, modeling, and evaluation are critical for natural listening behavior.

Online video understanding requires models to robustly encode, store, and retrieve information from a continuous stream to support accurate question answering. Existing approaches rely on key--value caching to accumulate frame-level representations over time, but allocate only a limited token budget per frame, causing loss of fine-grained visual detail. We find that naively scaling this budget degrades retrieval quality and question-answering performance, as denser token streams introduce redundancy that disrupts query--frame similarity. To address this, we first introduce an adaptive selection strategy that reduces token redundancy while preserving local spatiotemporal information. Second, we propose a training-free retrieval ensemble that leverages external models to better identify relevant frames. Our method, \textbf{MemStream}, achieves +8.0\% on CG-Bench, +8.5\% on LVBench, and +2.4\% on VideoMME (Long) over ReKV with Qwen2.5-VL-7B. Furthermore, we show that MemStream consistently improves performance across several Video-LLMs for long-video understanding.

Visual geometry transformers have become powerful architectures for multi-view 3D reconstruction, enabling joint prediction of multiple 3D attributes in a feed-forward manner. However, their computational cost grows quadratically with the input sequence length due to the global attention layers inside these models. This limits both their scalability and efficiency. In this work, we address this challenge with a simple yet general strategy: restricting the number of key/value tokens that each query interacts with during global attention. To achieve effective token selection, we introduce a two-stage framework. First, an inter-frame selection step operates at the frame level to identify frames that should be preserved. Second, an intra-frame selection step further discards more redundant tokens within the selected frames. Our analysis highlights the advantage of a diversity-based strategy for inter-frame selection, which ensures broad coverage of the scene. For intra-frame selection, we show that layer-aware sparsification is necessary, with the selection process guided by the entropy of the global attention pattern. Our approach offers a superior speed-accuracy trade-off compared to existing solutions. Extensive experiments show that it accelerates visual geometry transformers by over 85% for scenes with 500 images while maintaining, or even improving, baseline performance, which hints that how our token selection strategy can play a crucial role in future applications of visual geometry transformers.

Retrieval-augmented generation (RAG) divides into two camps failing in complementary ways on broad, long-tailed questions: iterative agentic RAG re-queries a biased distribution and drifts on prominent aspects, while structural RAG commits to a hierarchy fixed upfront and cannot admit aspects it initially missed. We propose GBT-RAG, which unifies the two by transposing gradient boosting into the text domain: each round is a query decomposition tree fitted to its predecessors' residual gap. This raises two challenges mirroring the pillars of boosting---natural-language critiques are not quantifiable, and successive rounds drift rather than fit the residual. We address the first with a text gradient, a structured residual in the retriever's embedding space whose support localizes missingness to specific leaves; we address the second by growing each tree only over under-covered leaves and admitting each rewrite through a monotone-improvement gate. Across standard benchmarks, GBT-RAG consistently outperforms the strongest iterative and structural baselines, with the largest gains on the long-tailed regime that motivated the design.

We introduce Gram-Calibrated Anchoring (GCA) for class-incremental learning (CIL) with pre-trained Vision Transformers. GCA rests on the observation that, under continual low-rank adaptation, anchor prototypes---class means extracted once from the frozen pre-trained backbone---are effective substitutes for prototypes recomputed from the adapted model; a formal stability analysis bounds decision-boundary shifts in terms of feature-level prototype drift, while empirical LoRA perturbation measurements support the small-drift regime. This motivates fixing the classification head to anchor prototypes computed upon each task's arrival rather than recalibrating prototypes after training. The fixed anchor geometry then enables Gram-Compensated Inference (GCI), which applies the regularized inverse of the anchor Gram matrix to deconvolve inter-class correlation that standard cosine classification ignores, and normalizes each coordinate by its estimation standard deviation for fair cross-class comparison. The resulting method trains a single continually evolved LoRA module with no exemplar storage or expanding adapter pools. Experiments on ImageNet-R, ImageNet-A, CIFAR-100, and CUB-200 with multiple backbones show that GCA achieves competitive accuracy across four CIL benchmarks while significantly reducing training cost over recent methods.

Deep networks achieve high realism in texture synthesis but often rely on opaque, over-parameterized representations that have limited direct connection to biological vision. Neuroscience-informed pyramid models are interpretable, compact, and biologically grounded, but rely on small, hand-curated statistics sets that are difficult to generalize or extend. We introduce GramStatTexNet, a hybrid analysis-by-synthesis framework that combines the multi-scale Gabor filter structure of classical models with the flexibility of Gramian-based correlations, organized into structured filter families. This structured representation enables analyses that are challenging for deep-feature methods: each statistic carries a named filter-family identity, allowing us to decompose synthesis fidelity into per-family contributions and to learn an interpretable, low-dimensional embedding that preserves perceptual texture content. We compare our framework to classical and deep texture-synthesis baselines across diverse texture categories, and on multiple perceptual metrics, highlighting the compactness and interpretability advantages of this hybrid representation, alongside competitive synthesis quality. This same framework extends naturally to peripheral textures via spatial pooling, and to dynamic textures using a spatiotemporal filter bank. GramStatTexNet provides a unified, interpretable framework for analyzing and modeling visual information with natural extensions across space, time, and eccentricity.

Tree-based Monte-Carlo Tree Search (MCTS) duplicates the same state when it is reached through different trajectories, which can waste simulations in stochastic MDPs. We introduce Graph-Based Stochastic-Power-UCT (GS-Power-UCT), which shares states reached at the same planning depth while keeping separate values for states reached at different depths. This design applies to general stochastic MDPs, including problems with cycles. We prove that for a fixed planning horizon, the root estimate converges to the finite-horizon value at rate $O(n^{-1/2})$, matching tree-based Stochastic-Power-UCT while reusing samples across shared states. We also study two full-state variants: GS-Power-UCT-F, which stores one node per physical state to increase sample sharing but may mix values from different remaining horizons, and GS-Power-UCT-F$^+$, which uses an adaptive horizon to control this bias. The latter converges to $V^{\star}(s_0)$, the optimal infinite-horizon discounted value at the root state $s_0$, when the remaining cross-depth gap vanishes. Experiments on stochastic planning benchmarks show improved sample efficiency over tree-based and graph-based baselines.

Long-term memory construction for LLM agents is hindered by two mismatches. The first is a retrieval-objective mismatch: existing systems usually retrieve memories for writing with a single holistic query derived from the current input. However, memory construction depends on several complementary aspects of the input, and a single query is insufficient to capture these editing needs. The second is an optimization-granularity mismatch: long-horizon memory learning is commonly optimized over full trajectories, whereas memory construction proceeds through incremental state transitions. Effective supervision should therefore assess how each local edit changes the memory state. We propose GraphMemRL, a graph-based memory construction framework that addresses both mismatches. GraphMemRL introduces edit-oriented query decomposition to retrieve complementary context for memory writing. It also uses action-native state-transition supervision to directly evaluate whether each graph edit improves the evolving memory state. To make long-horizon learning more tractable, GraphMemRL reformulates memory construction as factorized segment-wise optimization. The agent updates a persistent memory graph across segments under limited context budgets, preserving globally accumulated structure without full-history training. Experiments on long-memory benchmarks show that GraphMemRL improves question answering and produces better connected memory structures with more controlled memory growth. The gains are especially strong on multi-hop, temporal, and knowledge-update tasks.

Understanding social interactions requires reasoning over subtle non-verbal cues, yet current multimodal large language models (MLLMs) often fail to identify who interacts with whom in multi-person videos. We introduce GRASP, a large-scale social reasoning dataset that connects high-level social QA with fine-grained gaze and deictic gesture events. GRASP contains 290K question-answer pairs over 46K videos totaling 749 hours, organized by a 16-category taxonomy spanning gaze, gesture, and joint gaze-gesture reasoning, together with GRASP-Bench for evaluation. Unlike prior resources that focus on either isolated cues or high-level social QA, GRASP builds questions from identity-consistent gaze trajectories, deictic gestures, and their joint compositions into social events. Moreover, we propose Social Grounding Reward (SGR), a learning signal that uses these social events to encourage models to reason about the participants involved in each interaction. Experiments show that SGR improves performance on GRASP-Bench while maintaining zero-shot performance on related social video QA benchmarks.

Autoregressive decoding with LLMs is primarily bottlenecked by GPU memory bandwidth, especially in edge-computing settings. While quantization is essential for mitigating this bottleneck, most existing methods treat inference as uniform process and fail to account for the asymmetry between the compute-bound prefill stage and the memory-bound decoding stage. We propose GRINQH (GRaded INput-based Quantization Hierarchy), a weight-only post-training quantization framework that accelerates decoding by unifying quantization and sparsification. GRINQH leverages activation magnitudes as a proxy for computational importance to dynamically assign weight channels to different precision levels, enabling flexible average bit widths during decoding. Evaluated on Llama3 and Qwen3 models, GRINQH outperforms state-of-the-art fixed- and mixed-precision baselines at comparable 3- and 4-bit settings, even enabling effective 2-bit generation. We experimentally verify theoretical speedups by leveraging a hierarchical nested memory layout for multi-precision storage in custom GPU kernels. Together, GRINQH establishes a new state-of-the-art Pareto frontier for LLM generation, enabling a dynamic trade-off between generation quality and inference speed.

Hallucinations in large language models (LLMs) that are plausible-sounding but factually incorrect or unsupported pose a major challenge for deploying these models in high-stakes applications such as medical diagnosis, legal reasoning, and knowledge-based question answering. Existing detection methods primarily rely on either generating multiple responses to check consistency or supervised training on labeled hallucinations. Multiple-response approaches tend to be computationally expensive and may be less practical for real-time scenarios, while supervised methods are limited to known hallucination types and cannot generalize to unseen cases. To address these limitations, we propose an efficient unsupervised hallucination detection method. Our approach measures the embedding discrepancy between the LLM’s response considered independently and the response contextualized with the question. A large discrepancy indicates that the question significantly influences the response, suggesting it is grounded, whereas a small discrepancy signals a higher likelihood of hallucination. This method is efficient, interpretable, and does not require labeled hallucinations. Extensive experiments across multiple tasks demonstrate that our approach consistently achieves superior or comparable detection performance while reducing the detection cost compared to the multiple-response methods.

3D affordance grounding aims to highlight the actionable regions on 3D objects, which is crucial for embodied AI. While previous research primarily leverages static cues from language or single images, which often lack the rich interaction context necessary to accurately locate functional zones. To alleviate this predicament, we collect a comprehensive video-based 3D affordance dataset, VIDA, which contains 38K human-object-interaction (HOI) videos covering 16 affordance types, 38 object categories, and 22K point clouds. Based on VIDA, we propose a strong baseline: VideoAfford, a unified framework that extends Multimodal Large Language Models (MLLMs) with fine-grained affordance segmentation capabilities. VideoAfford incorporates a latent action encoder to distill dynamic interaction knowledge from demonstration videos and a spatial-aware loss to encourage geometrically consistent affordance predictions. Extensive experiments on VIDA show that VideoAfford significantly outperforms strong baselines in both seen and unseen settings, demonstrating its effectiveness in video-driven 3D affordance reasoning and open-world generalization. The dataset and code will be released soon to facilitate future research in this area.

Long-horizon LLM agents trained with outcome-based RL exhibit a systematic failure mode: reasoning diverges from environmental observations at some step, and the error compounds across subsequent turns. We find that 74\% of failed trajectories contain reasoning–observation inconsistency, and once it appears, 62\% of subsequent turns remain inconsistent. Process rewards reduce but do not eliminate this: under free-form reasoning, agents learn to avoid penalties without genuinely tracking task progress or anticipating action outcomes. We propose ARC (Anchored Reasoning with Commitment), which requires the agent to make two explicit commitments at each step: a global progress statement that reconciles the agent's understanding of task state with accumulated observations, and a local outcome prediction that must be borne out by the next observation. Together, these commitments anchor the agent's reasoning to the actual environment state across turns, and provide well-defined targets for process supervision. To prevent interference between sparse outcome and dense process signals, we decouple their advantages via independent group normalization. On ALFWorld, WebShop, and WebArena, ARC improves over the strongest baseline by 6.4, 3.5, and 9.6 points respectively.

Group distributionally robust optimization (GDRO) aims to learn models that perform well across $m$ distributions simultaneously. Existing methods either (i) query $m$ samples per round, which imposes strict implementation requirements, or (ii) query only a single sample per round, which requires many rounds to converge and therefore leads to long runtime. To avoid these limitations, we introduce a novel setting called \emph{flexible sample queries} (FSQ), which allows the number of samples that can be queried to vary across rounds. Under the FSQ setting, we cast GDRO as a two-player game: one performs non-oblivious online convex optimization, while the other tackles a non-oblivious prediction with limited advice (PLA) problem. In this game, the first player updates its decision using follow-the-regularized-leader with stochastic gradients. For the second player, we develop a new PLA method equipped with an adaptive exploration strategy that can leverage an arbitrary per-round sample size. We then establish the *first* high-probability regret bound for non-oblivious PLA. By integrating these components, we develop an anytime GDRO algorithm that supports FSQ, and derive a high-probability optimization error bound of $O\left(\frac{1}{t}\sqrt{\sum_{j=1}^t \frac{m}{r_j}\log m}\right)$, where $r_j$ is the sample size at round $j$. This result demonstrates that the optimization error decreases as the per-round sample sizes increase, and implies the same near-optimal sample complexity of $O(m\log (m)/\epsilon^2)$ for any fixed sample size $r\in[m]$.

This paper proposes a guaranteed defense method for large language models (LLMs) to safeguard against jailbreaking attacks. Drawing inspiration from the denoised-smoothing approach in the adversarial defense domain, we propose a novel smoothing-based defense method, termed Disrupt-and-Rectify Smoothing (DR-Smoothing). Specifically, we integrate a two-stage prompt processing scheme—first disrupting the input prompt, then rectifying it—into the conventional smoothing defense framework. This disrupt-and-rectify approach improves upon previous disrupt-only approaches by restoring out-of-distribution disrupted prompts to an in-distribution form, thereby reducing the risk of unpredictable LLM behavior. In addition, this two-stage scheme offers a distinct advantage in striking a balance between harmlessness and helpfulness in jailbreaking defense. Notably, we present a theoretical analysis for generic smoothing framework, offering a tight bound for the defense success probability and the requirements on the disruption strength. Our approach can defend against both token-level and prompt-level jailbreaking attacks, under both established and adaptive attacking scenarios. Extensive experiments demonstrate that our approach surpasses current state-of-the-art defense methods in terms of both harmlessness and helpfulness.

Recent advances in visual storytelling have leveraged powerful text-to-image diffusion models to generate sequences of images featuring consistent subjects and styles. While such methods achieve impressive identity preservation across the sequence, they often produce scenes with limited structural diversity---images that are overly similar in subject pose and spatial layout. Existing evaluation protocols primarily measure identity consistency using metrics such as DreamSim, which can inadvertently conflate identity preservation with subject pose. The inability to independently evaluate structural diversity inhibits progress on identity preserving image generation methods. We address this limitation by introducing a new metric based on the Gromov Wasserstein distance to quantify structural diversity independently of identity. Our metric provides an interpretable and generalizable measure of spatial variation across generated images, enabling a more balanced assessment of identity consistency and pose diversity. Using this framework, we systematically evaluate several recent visual storytelling and subject-consistent generation methods. Furthermore, we evaluate the effect of latent initialization methods on structural diversity in several popular text-to-image models.

Aligning autonomous agents with human intent remains a central challenge in modern AI. A key manifestation of this challenge is reward hacking, whereby agents appear successful under the evaluation signal while violating the intended objective. Reward hacking has been observed across a wide range of settings, yet methods for reliably measuring it at scale remain lacking. In this work, we introduce a new evaluation paradigm for measuring reward hacking. Whereas prior studies have primarily analyzed it post hoc by inspecting agent trajectories, we instead embed detectable reward hacking opportunities directly into environments. This makes their exploitation verifiable by design, enabling deterministic and automated measurement of whether and how agents exploit such vulnerabilities. We instantiate this approach in TextArena and release Hack-Verifiable TextArena, a testbed in which reward hacking can be measured reliably. Using this benchmark, we analyze reward hacking behavior across language models in diverse environments and settings.

Selective unlearning in large language models is increasingly necessary for safety and compliance, yet harmful or memorized content is often latent, entangled, and hard to specify as a clean ``forget set.'' We aim to close this gap by identifying what to forget from the model’s own failure modes rather than relying solely on explicit supervision. We propose Hallucination-Guided Unlearning (HGU), which treats hallucinations as diagnostic signals: it localizes unsupported spans via evidence and internal consensus, traces a sparse set of high-responsibility MLP neurons, and edits them with compact gates under a utility-preserving KL constraint. Across TOFU, HGU achieves the best MU--FE trade-off with Avg of 0.7221 (forget05) and 0.7699 (forget10), and it also attains the top SepS H-Avg (0.4382/0.4715) under mixed forget/retain prompts; on WMDP auxiliary, it yields the strongest forgetting (e.g., 14.2 on Economics) while improving Retain (52.7) and MMLU (57.2). These results suggest that reframing hallucinations as actionable traces enables targeted, post-hoc unlearning that better balances forgetting efficacy and general utility without full retraining.

Federated Learning (FL) with Low-Rank Adaptation (LoRA) enables privacy-preserving collaborative fine-tuning of foundation models. Existing federated LoRA studies often assume homogeneous ranks and lack principled rank allocation, leading to underfitting or overfitting under data and resource heterogeneity. Although later heterogeneous-rank federated LoRA methods relax the homogeneous-rank assumption, they still ignore resource-adaptive rank allocation across clients, while their aggregation and distribution strategies suffer from structural limitations, including rank collapse in naive truncation and the loss of locally relevant update signals caused by objective mismatch in SVD truncation. To address these issues, we propose HALO, a heterogeneous federated LoRA framework that integrates adaptive rank allocation with client-aware aggregation and distribution. Under the neural tangent kernel regime, we derive a sufficient rank condition to guide client-specific rank allocation from local data geometry and resource constraints. We further propose Client-Aware Projection (CAP), which builds a product-summed reference and redistributes rank-compatible balanced LoRA factors through projections onto client-induced subspaces, improving generalization via cross-client information while preserving locally relevant update signals. We show that CAP controls the client-loss-relevant truncation residual under the NTK approximation without additional communication overhead, which stabilizes local training and improves performance. Extensive experiments demonstrate the effectiveness of HALO. Our code is available at https://anonymous.4open.science/r/artifact2026-4174.

Supervised post-training, including supervised fine-tuning (SFT) and knowledge distillation (KD), is essential for adapting large language models (LLMs) to specialized domains. However, traditional supervised post-training methods often suffer from catastrophic forgetting, training instability, and prohibitive computational costs. While layer-wise strategies have emerged as efficient alternatives, they introduce a representation bottleneck, where constrained updates in early stages limit feature diversity and impair final generalization. To resolve these issues, we develop HALO (Homotopy-Augmented Layer Optimization), a homotopy-driven layer-wise strategy for LLM post-training. By introducing a continuous homotopy parameter and a sequence of monotonically increasing auxiliary functions, we formulate the post-training process as a continuous deformation of the optimization problem that progressively incorporates layers into the trainable set. Specifically, HALO unfreezes the model from the output back to the input in a structured sequence. We theoretically prove that this approach establishes a differentiable solution path, ensuring a stable transition from a restricted optimization state to a fully converged LLM. Extensive experiments across diverse model families and sizes demonstrate that HALO consistently enhances performance and generalization. Notably, HALO exhibits strong robustness in challenging scenarios, especially in low-resource settings, while significantly reducing convergence time and iteration counts. Our results position HALO as an effective and accessible framework for high-performing LLM specialization.

Physical reasoning about humans in videos has largely focused on everyday interactions, where contact, pressure, pose, and motion support embodied perception and contact-aware applications. Still, short and forceful properties like impact events remain relatively underexplored, despite their relevance to broader understanding of how abrupt physical events affect the human body. To fill this gap, we introduce human-centric impact localization, a new task that localizes target physical impacts temporally within a video and spatially on the human body. We further present HAPACT, a HumAn-centric imPACT dataset and benchmark built from movie clips, containing over 2.8K video-query pairs with more than 20K fine-grained human-annotated frames. We built HAPACT around viewer-grounded annotation, text-conditioned localization, and supervision at frame-level temporal and SMPL-X-based spatial granularity. To demonstrate the effectiveness of our benchmark, we conduct extensive evaluations of existing methods, employing video moment retrieval models for temporal localization and human-scene contact models for spatial localization. For the spatial task, we propose a text-conditioned baseline by reformulating localization as query-guided body-region selection. Benchmark results show that physical impact localization remains challenging for existing methods, motivating HAPACT as a dedicated benchmark to support physical impact reasoning in videos.

LLMs often struggle with memory-constrained deployment on consumer-grade hardware due to their massive parameter sizes. While existing solutions such as model compression and offloading improve deployment feasibility, they often suffer from substantial accuracy degradation or severe throughput bottlenecks. Recent error compensation methods recover accuracy through auxiliary LoRA-style branches, and we observe that these branches are inherently amenable to offloading: they require substantial parameter storage but access only a small subset of compensation parameters during each inference step. Motivated by this opportunity, we propose HCInfer, a heterogeneous inference system that offloads residual compensation to the CPU while executing the compressed backbone on the GPU, and further introduces an asynchronous compensation pipeline and sensitivity-aware dynamic rank allocation to hide compensation overhead and maximize accuracy recovery. Experimental results show that HCInfer achieves a maximum accuracy improvement of 5.2% on downstream tasks compared to compression model and sustaining a maximum speedup of 10.4x compared to full-precision model.

Large vision-language models (LVLMs) can process interleaved text and multiple images in a single context, but how they internally identify the image relevant to a question remains poorly understood. We study this grounding step as in-context image retrieval and introduce ICIR-MCQ, a controlled probe for analyzing image retrieval at attention-head granularity. Using each head's attention mass over image-token spans, we first identify sparse pointer heads that reliably point to the target image. These heads transfer beyond the controlled probe: a single pointer head outperforms external vision-language retrievers on a naturalistic multi-image retrieval benchmark, showing that LVLMs contain a strong internal image retrieval signal. However, attention alone only shows where a head reads from, not what information its output passes on to later layers. We therefore introduce a complementary output-side analysis. For each head, we train a lightweight image retriever on its value-weighted output and use it to identify writer heads, whose outputs contain enough information to identify the target image. Surprisingly, the pointer and writer criteria select substantially different head sets, with only 35 to 44 heads overlapping among the top-100 heads under each criterion. Targeted set-partition ablations show that the heads most important for standard multi-image inference are concentrated in this pointer-writer intersection, while heads selected by only one criterion contribute little beyond random-head controls. These results show that attention-based pointing alone is systematically incomplete for identifying image retrieval heads in LVLMs. Causally relevant retrieval heads must also write target-image information to the residual stream, not merely point to it.

Audio-visual large language models (AV-LLMs) have made strong progress in multimodal understanding, but they typically treat audio as monaural semantic content and therefore struggle to reason about where sounds originate and how they relate to the visual scene. Recent spatial audio-visual studies address parts of this problem, but often focus on specific abilities, such as spatial correspondence or direction and distance reasoning. In this paper, we propose HLR-AVSceneQA, a comprehensive benchmark for spatial audio-visual scene understanding. HLR-AVSceneQA evaluates whether models can jointly hear, localize, and reason by recognizing what is heard, grounding where it comes from in egocentric and allocentric views, and inferring how sound sources relate to visible, hidden, or nearby objects. We further introduce HLR-LLM, which extends a strong audio-visual foundation model with a binaural spatial audio branch. HLR-LLM is trained with a three-stage curriculum, applying chain-of-thought supervision in the final stage to help the model learn structured multi-step spatial reasoning. Experiments show that existing AV-LLMs remain limited in spatial grounding and relational reasoning, while HLR-LLM substantially improves spatial audio-visual scene understanding without sacrificing semantic audio-visual perception.

Individual-level genotype data underpin genetic discovery, disease-risk modeling, and therapeutic development, yet many critical settings remain data-limited, including underrepresented ancestries, rare or imbalanced disease cohorts, and restricted-access biobanks. Conditional synthetic genome generation would support privacy-aware benchmarking, simulation, and data augmentation, but requires learning an ultra-high-dimensional discrete distribution over hundreds of thousands to tens of millions of correlated single-nucleotide polymorphisms (SNPs), or sites of DNA variation. We introduce $\mathbf{HELICS}$: $\mathbf{H}$igh-fidelity g$\mathbf{E}$neration via $\mathbf{L}$atent $\mathbf{I}$nterpolant $\mathbf{C}$onditional $\mathbf{S}$ynthesis, a phasing-free latent flow matching framework for genome-wide SNP-level generation from biobank-scale genotype data. HELICS tokenizes chromosomes with convolutional autoencoders, reducing $>$$600{\rm K}$ SNPs to $4{,}096$ continuous latent tokens, and trains a transformer parameterized conditional flow matching model over the concatenated multi-chromosome latent representation. Trained on roughly $428\rm K$ genomes in the UK Biobank, HELICS generates genome-wide synthetic cohorts without haplotype phasing and achieves stronger fidelity to held-out real genotypes than existing haplotype- and genotype-based simulators, including the highest validation allele-frequency agreement ($R^2=0.992$) and superior preservation of local and long-range SNP correlation structure. Conditioning on ancestry and genetic liability summaries enables targeted generation across genetic risk profiles; in perturbation analyses, by increasing the conditioned disease trait liability, HELICS induces SNP-level dosage changes in generated genomes that are highly correlated with published association weights. HELICS provides a scalable generative modeling framework for SNP-level genome data and a path toward conditional synthetic cohorts for data-limited genetic studies.

Zeroth-order stochastic optimization is a fundamental formulation that arises in real-world design problems where gradients are inaccessible. A central challenge in this pursuit is to design gradient estimators and optimization algorithms under noisy, function-only feedback that exploit local Hessian geometry to achieve optimal sample efficiency. We introduce the Spectrally Grouped Estimator (SGE), a novel gradient estimator that samples over a non-convex union of sphere sections, and utilize it to build an algorithm that achieves order-wise improved Hessian-dependent simple regrets over second-order smooth, strongly convex functions compared to conventional convex-sets sampling baseline methods. We complement these results with the first tight analyses of the baseline schemes, revealing a shared performance bottleneck and thus emphasizing the necessity of non-convex sampling for optimality. We further establish matching converse bounds that tightly characterize the optimal sample complexities within general subclasses of functions sharing the same Hessian at the global minimum, thereby proving the universal optimality of SGE with respect to the Hessian-dependent rates. This fully resolves an open conjecture from the preliminary version of this work. This fully resolves an open conjecture in the preliminary version of this work.

Subgraph heterogeneity is a critical challenge in federated learning, significantly impacting model performance. Data-free knowledge distillation overcomes the limitation of sharing private label distributions in federated learning, yet existing research lacks in-depth exploration of subgraph heterogeneity. Our work builds upon the fact that node and structural variations in heterogeneous graphs cause significant disparities in the reliability of knowledge from local graph neural networks. We propose a structure-aware bidirectional data-free federated distillation technique, where the generator and global model engage in an adversarial training process to ensure reliable bidirectional knowledge transfer between local and global models. Extensive experiments across multiple public datasets demonstrate the model's effectiveness.

Adapting pretrained diffusion models to downstream objectives such as inverse problems often requires expensive test-time guidance or optimization. We propose a principled framework for generating high-quality reward-aligned samples at substantially reduced inference cost. Our approach formulates test-time adaptation as a hierarchical variational model, where control is amortized into a lightweight yet expressive stochastic policy. This formulation naturally supports few-step diffusion sampling: large step sizes enable fast inference, while the learned policy maintains sample quality by providing structured per-step control. The resulting fully amortized sampler achieves a strong quality--speed tradeoff, matching or exceeding recent test-time scaling baselines while requiring significantly less compute. For example, on $4\times$ super-resolution, our method achieves better perceptual quality with more than 5$\times$ faster inference compared to the best-performing baseline. We further extend our approach to a semi-amortized regime that combines cheap amortized proposals with limited test-time optimization, achieving state-of-the-art perceptual quality across several challenging inverse problems.

Planning with world models enables test-time adaptation for embodied agents, offering superior generalization over direct policy learning. However, long-horizon tasks remain hindered by cumulative errors and search complexity. Existing hierarchical methods often use rigid macro-actions or latent skills, leading to brittle representations and limited fine-grained control. We introduce Implicit-HWM, a hierarchical world model that decouples state-space reachability from local control. Our framework utilizes a high-level model to sample feasible transitions across the state manifold and a low-level inverse dynamics solver to ground these transitions into precise actions. This allows the planner to identify reachable subgoals without being constrained by low-level execution details. By prioritizing local physical feasibility over global behavioral patterns, Implicit-HWM synthesizes novel long-horizon plans from short segments, transcending the specific behavioral modes seen during training. We demonstrate its effectiveness across three navigation and manipulation suites, achieving a 38.5% improvement over state-of-the-art policy and diffusion-based planners. Crucially, by leveraging this hierarchical structure to mitigate compounding errors, our approach yields a 120% performance gain over single-level world model MPC, particularly in complex, unseen configurations for long-horizon tasks.

Existing multimodal large language models (MLLMs) typically formulate object detection as a conditional sequence generation task. Under this paradigm, spatial locations are commonly represented by either textual coordinates or special quantized learnable tokens. However, textual coordinates are tokenized as unrelated symbols, spatially adjacent points (e.g., $x=199$ vs. $y=200$) yield entirely different tokens, so the cross-entropy loss is misaligned with true spatial distance. Special quantized location tokens, on the other hand, inevitably introduce quantization error. Reducing this error requires linearly more tokens, but a larger vocabulary is harder to train and converges slower. We argue that both limitations stem from the formulation of spatial representation. To address this issue, we propose a simple yet novel localization representation method, HiLoc, which predicts locations in a coarse-to-fine manner with special tokens: first a grid index for coarse localization, then a cell index to refine the position within the grid. Under this paradigm, accuracy can be scaled up by simply adding the hierarchy level. We also build HiRef-500K, a dataset consisting negative and complex referring samples to strengthen rejection and complex referring ability. Experiments show that our method achieves strong performance on both detection and grounding benchmarks, especially on small objects, under high IOU thresholds and complex referring tasks.

Offline goal-conditioned reinforcement learning methods have shown promise for reach-avoid tasks, where an agent must reach a target state while avoiding undesirable regions of the state space. Existing approaches typically encode avoid-region information into an augmented state space and cost function, which prevents flexible, dynamic specification of novel avoid-region information at evaluation time. They also rely heavily on meticulously designed reward and cost functions, limiting transferability to novel environments and specifications. We introduce RADT, a decision transformer approach for offline, reward-free, goal-conditioned, avoid-region-conditioned RL. RADT encodes goals and avoid-regions directly as prompt tokens, allowing any number of avoid-regions of arbitrary size to be specified at evaluation time. Using only suboptimal offline trajectories from a random policy, RADT can learn reach-avoid behavior in a completely data-driven manner using our novel avoid-region hindsight relabeling approach, without the need for reward/cost supervision. We benchmark RADT against existing offline goal-conditioned RL models across 17 tasks, environments, and experimental settings. RADT generalizes in a zero-shot manner to out-of-distribution avoid-region sizes and counts, outperforming baselines that require retraining. In one such zero-shot setting, RADT achieves 35.7% improvement in normalized cost over the best retrained baseline while maintaining high goal-reaching success. We also apply RADT to cell reprogramming in biology, demonstrating its versatility.

Existing conformal prediction methods for time-to-event outcomes leverage only baseline covariates, producing prediction intervals that are insufficiently informative to facilitate decision making. We propose History-Aware Prediction Sets (HAPS), a conformal framework that constructs prediction sets for individual event times using covariate histories observed up to a decision time, targeting coverage among individuals who have survived to this time. HAPS handles right censoring adjusted for time-varying confounders via inverse probability of censoring weighting. When the censoring weights are consistently estimated, it achieves PAAC (probably asymptotically approximately correct) coverage among survivors. We further propose two doubly robust extensions of HAPS to weaken reliance on consistent estimation of the censoring distribution. In simulations, HAPS and its extensions reduce median prediction interval length by up to 75\% relative to baseline comparators while maintaining close to nominal coverage. On two public benchmark data sets, HAPS reduces the median interval length by up to 60\% for predictions at year 5, compared to the baseline comparators.

Prior Laplacian-based option-discovery methods build options from 0-form Laplacians defined on nodes, where the resulting eigenvectors induce routes toward isolated boundaries or extrema, thereby skewing visitation toward specific states. In contrast, kernel eigenvectors of the 1-form Hodge Laplacian represent interpretable, topology-aware edge flows and naturally capture cyclic behaviour known as harmonic flows. These kernel eigenvectors are particularly attractive because they define zero-divergence edge flows, offering a natural solution for non-isolated visitation to sources and sinks in regions in which they are active. Despite this appeal, harmonic flows have not previously been explored as a basis for option discovery in reinforcement learning. This is what we do here. We first introduce harmonic-flow skills and show that, on state graphs with tree-like “dead-end” appendages, purely harmonic flows can become nearly inactive in those regions, which can limit coverage. To address this limitation, we propose quasi-harmonic flow options: skill-inducing edge flows that preserve harmonic cycles where harmonic activity is strong, while injecting drift into appendages using the lowest-magnitude non-harmonic eigenvector of the Hodge Laplacian, a low divergence solution, where harmonic activity is weak. Our approach is eigenvector-only, relying solely on spectral components of the Hodge Laplacian. We compare quasi-harmonic flow options against a harmonic-flow options baseline and state-of-the-art skill-discovery methods from the literature, demonstrating empirical gains and analytical improvements on hitting-time-based metrics.

Human-object-centric video personalization (HOCVP) is a core task within subject-driven video generation. However, existing methods suffer from two key limitations. First, most approaches focusing on $\textit{inter-subject}$ personalization still struggle to strike a balance between high subject fidelity and accurate interaction patterns between humans and diverse objects, especially when objects represent abstract concepts such as logos. Second, while $\textit{intra-subject}$ references ($\textit{e.g.}$, OCR maps, multi-view inputs) are expected to enhance subject fidelity, most existing works lack mechanisms to understand such latent correspondence. To address both challenges, we propose HOMIE, an HOCVP framework that tackles both inter- and intra-subject input settings in a unified manner. Compared to previous approaches, HOMIE proposes a better MLLM integration strategy to extract knowledge of reference-level relationships without compromising the controllability of text encoders or incurring costly re-alignment. Specifically, we introduce global multimodal guidance within self-attention to better align MLLM-derived semantic features with VAE tokens. Furthermore, we propose modality-reference embedding to differentiate tokens from MLLM features and VAE tokens and associate intra-subject reference image tokens. Extensive experiments validate that our method achieves state-of-the-art performance across various HOCVP tasks. Anonymous project page with videos is available at https://homie-demo.github.io/.

We study signSGD in linear regression with diagonal features and power-law spectra, a toy but expressive setting where the effect of coordinate-wise sign normalization can be analyzed sharply. Specifically, we establish a functional scaling law (FSL) for signSGD under general learning rate schedules, which decomposes the loss dynamics into two components: learning of the target signal and noise accumulation. Compared with SGD, signSGD learns the signal faster but exhibits slower noise forgetting. This signal-noise tradeoff yields a regime-dependent comparison between signSGD and SGD: signSGD achieves better data-scaling efficiency in hard-task regimes, while they perform similarly in easy-task regimes. The key mechanism is that gradient noise is {\it curvature-aligned}: its directional variance is of the same order as the directional curvature. As a result, sign operation effectively acts as a \(\diag(\bH)^{-1/2}\) preconditioner in the noise-dominated regime, where $\bH$ is the Hessian matrix. Moreover, we demonstrate that the same square-root preconditioning mechanism can arise in RMSprop and Adam, where coordinate-wise normalizers essentially track the coordinate curvatures. Synthetic experiments validate the proposed mechanism and the predicted scaling behavior. Language model experiments further suggest that, beyond the controlled setting, the resulting FSL can serve as a surrogate model for fitting and predicting practical training loss curves.

Despite their success in image generation, diffusion models can memorize training data, raising serious privacy and copyright concerns. Although prior work has identified empirical factors associated with memorization, such as text guidance, prompt specificity, and duplicated training examples, the mechanism by which memorization is triggered and propagated during denoising remains unclear. In this paper, we provide a theoretical explanation of how memorization occurs in text-to-image diffusion models. We show that conditional overfitting causes the conditional posterior mean to collapse to the memorized training latent, while the unconditional posterior remains close to the zero-centered data mean at the first denoising step. As a result, classifier-free guidance overestimates the clean prediction and injects an amplified memorized signal into the reverse trajectory. We further show that this signal propagates across timesteps because each guided latent is fed back into the denoising network, allowing the conditional branch to indirectly shift the unconditional branch. To characterize this process, we introduce the posterior lag, the discrepancy between conditional and unconditional posterior means. For normal prompts, this lag decreases as the two posteriors synchronize during denoising. For memorized prompts, the conditional posterior commits early to a specific training latent, while the unconditional posterior catches up only later, producing a distinctive rise-and-fall lag pattern. Our analysis provides a mechanism-level explanation of memorization and clarifies why text guidance and early denoising steps play a central role in memorized generation.

Large language model (LLM)-powered assistants have recently integrated memory mechanisms that record user preferences, leading to more personalized and user-aligned responses. However, irrelevant personalized memories are often introduced into the context, interfering with the LLM's intent understanding. To comprehensively investigate the dual effects of personalization, we develop \textsc{RPEval}, a benchmark comprising a personalized intent reasoning dataset and a multi-granularity evaluation protocol. \textsc{RPEval} reveals the widespread phenomenon of irrational personalization in existing LLMs and, through error pattern analysis, illustrates its negative impact on user experience. Finally, we introduce \textsc{RP-Reasoner}, which treats memory utilization as a pragmatic reasoning process, enabling the selective integration of personalized information. Experimental results demonstrate that our method significantly outperforms carefully designed baselines on \textsc{RPEval}, and resolves 80\% of the bad cases observed in a large-scale commercial personalized assistant, highlighting the potential of pragmatic reasoning to mitigate irrational personalization. Our data is available at \url{https://anonymous.4open.science/r/RPEval-E4B0}.

Post-training pruning reduces the memory and computation costs of large language models (LLMs), and N:M semi-structured sparsity provides a hardware-friendly pruning pattern for efficient inference. However, pruned models are often evaluated mainly by perplexity and average downstream accuracy, which are useful aggregate metrics but do not fully explain how pruning changes individual candidate-level decisions. We evaluate magnitude pruning, Wanda, SparseGPT, Wanda++, and our proposed MR-GOBS across multiple model families and sparsity settings. We further conduct a paired decision-level evaluation that measures true-label confidence, answer margin, prediction flips, candidate-level distribution shift, and near-boundary vulnerability. Our results show that pruning weakens decision reliability by lowering confidence in the correct answer, shrinking answer margins, and increasing correct-to-wrong flips, especially on near-boundary examples. To examine whether reconstruction-oriented improvements reliably improve decision reliability, we revisit SparseGPT through its local least-squares reconstruction formulation and refine its N:M mask-selection step. MR-GOBS uses GroupOBS, a set-level OBS reconstruction cost for jointly pruning a group of weights, to score each legal local N:M mask. It then selects masks using a minimax-regret criterion over multiple calibration views. Although MR-GOBS often improves average perplexity over SparseGPT, these gains do not always translate into better downstream accuracy or decision-level metrics. Our findings suggest that reliable evaluation of pruned LLMs requires both aggregate performance metrics and decision-level diagnostics.

As Vision-Language Models (VLMs) are increasingly deployed as autonomous cognitive cores for embodied assistants, evaluating their privacy awareness in physical environments becomes critical. Unlike digital chatbots, these agents operate in intimate spaces, such as homes and hospitals, where they possess the physical agency to observe and manipulate privacy-sensitive information and artifacts. However, current benchmarks remain limited to unimodal, text-based representations that cannot capture the demands of real-world settings. To bridge this gap, we present ImmersedPrivacy, an interactive audio-visual evaluation framework that simulates realistic physical environments using a Unity-based simulator. ImmersedPrivacy evaluates physically grounded privacy awareness across three progressive tiers that test a model's ability to identify sensitive items in cluttered scenes, adapt to shifting social contexts, and resolve conflicts between explicit commands and inferred privacy constraints. Our evaluation of 12 state-of-the-art models reveals consistent deficits. In cluttered scenes, all models exhibit monotonic performance decay as scene complexity grows due to perceptual deficit. When social context shifts, no model exceed 65% selection accuracy. Under conflicting commands, the best model gemini-3.1-pro perfectly balances task completion and privacy preservation in only 51% of cases. These findings reveal that current VLMs in the physical world suffer from perceptual fragility and fail to let their knowledge of privacy cues govern their situated behavior. Our code and data is available at https://anonymous.4open.science/r/immersed-privacy-review-3A3C.

Stopgrads are widely used in training machine learning models, but stopgrads can alter the gradient, stationary points and convergence guarantees of the original objective, which can make stopgrad training theoretically ungrounded. We introduce a \textit{stopgrad regression principle}, which identifies a general template for stopgrad objectives with a closed-form characterization of stationary points, and unifies stopgrad objectives for flow maps, reinforcement learning, and diffusion samplers. We provide theoretical grounding for optimizing stopgrad flow map objectives by characterizing stationary points and proving convergence results. Remarkably, we show that the learned flow map has a closed-form expression composing the initial flow map and the true flow map, under functional semi-gradient flow. We use our stopgrad regression principle to propose novel modified stopgrad placements for flow map objectives which speed up training by ${\sim}2.5\times$.

We measure how much one recurrence is worth to a looped (depth-recurrent) transformer, in equivalent unique parameters. From an iso-depth pretraining sweep across recurrence counts $r \in \{1, 2, 4, 8\}$ spanning ${\sim}50\times$ in training compute, we fit a joint scaling law $L = E + A\,(N_\text{once} + r^{\varphi} N_\text{rec})^{-\alpha} + B\,D^{-\beta}$ and measure a recurrence-equivalence exponent $\varphi = 0.46$. Intuitively, $\varphi$ tells us whether looping a block $r$ times is equivalent in validation loss to $r$ unique blocks of a non-looped model (full equivalence, $\varphi{=}1$) or to a single block run repeatedly with no capacity gain ($\varphi{=}0$). Our $\varphi = 0.46$ sits in between, so replacing unique blocks with shared recurrences increases validation loss at matched training compute. For example, at $r{=}4$ a 410M looped model performs on par with a 580M non-looped model, but incurs the training cost of a 1B non-looped one. We demonstrate the utility of $\varphi$ as a diagnostic tool on two case studies: commonly used truncated backpropagation lowers $\varphi$ to $0.38$, indicating that the loop mechanism is poorly trained under truncation, even though validation loss decreases. Conversely, hyperconnections raise $\varphi$ to $0.65$, a genuine capacity gain. Our method separates true loop improvements from training-side gains, a distinction raw validation loss cannot make.

Post-training is central to building scientific reasoning models, but its effects remain poorly understood in biology, where models integrate natural language with multimodal biological data. We study how post-training stages shape generalization, and when they improve performance versus induce over-specialization. Across genomics, transcriptomics, and proteomics, we train and evaluate more than 100 biological reasoning models under controlled variation in backbone, continued pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL), measuring both in-domain (ID) and out-of-domain (OOD) performance. We find that post-training induces stage-specific trade-offs rather than uniform gains. CPT improves downstream performance by aligning models with biological language. SFT consistently increases ID performance but causes OOD performance to peak early and decline as models fit the training distribution. RL, when applied to strong SFT checkpoints with aligned rewards, improves OOD performance and partially recovers generalization. These results show that scientific reasoning does not improve monotonically with additional supervision or compute. Instead, performance depends on how training stages are composed. Under fixed epoch-level budget, the best ID-OOD trade-off comes from combining minimal SFT with larger RL budgets and allocating adaptation capacity asymmetrically across stages.

Concept Bottleneck Models (CBMs) have become a popular approach to enable interpretability in neural networks by constraining classifier inputs to a set of human-understandable concepts. While effective, current models embed concepts in flat Euclidean space, treating them as independent, orthogonal dimensions. Concepts, however, are highly structured and organized in semantic hierarchies. To resolve this mismatch, we propose Hyperbolic Concept Bottleneck Models (HypCBM), a post-hoc framework that grounds the bottleneck in this structure by reformulating concept activation as asymmetric geometric containment in hyperbolic space. Rather than treating entailment cones as a pre-training penalty, we show they encode a natural test-time activation signal: the margin of inclusion within a concept's entailment cone yields sparse, hierarchy-aware activations without any additional supervision or learned modules. We further introduce an adaptive scaling law for hierarchically faithful interventions, propagating user corrections coherently through the concept tree. Empirically, HypCBM rivals post-hoc Euclidean models trained on 20$\times$ more data in sparse regimes required for human interpretability, with stronger hierarchical consistency and improved robustness to input corruptions.

Compute-optimal scaling laws have become the central organizing tool of language-model training, yet they have been measured exclusively under Euclidean readouts. We give the first compute-optimal scaling-law characterization of hyperbolic language modeling. Trained from scratch on OpenWebText across a paired 5×4 grid spanning 86M to 1.32B parameters and 66M to 8.4B tokens, our hyperbolic LM HypTip follows a markedly different scaling law than its Euclidean counterpart. Fitting the Hoffmann form L = E + A·N^(−α) + B·D^(−β) to each track yields a compute-optimal data-to-parameter ratio of D*/N* = 8.4 for the hyperbolic readout versus 20.8 for the Euclidean baseline. The hyperbolic readout therefore reallocates compute by 2.5× toward parameter-heavy training, winning 17 of 20 paired cells across the grid. This shift is mechanistically grounded: across all 20 trained checkpoints, the per-row Lorentzian radius of the LM head correlates with token log-frequency at Spearman ρ down to −0.83, and the strength of this Zipfian token hierarchy scales with training-token count rather than parameter count, indicating that the geometry-as-prior emerges causally during training rather than being read off a frozen Euclidean checkpoint as in prior post-hoc analyses. Held-out evaluation on WikiText-103 perplexity, LAMBADA, and PIQA preserves the same advantage, and a cross-architecture replication on GPT-2 and DeepSeek-V3 confirms that LM-head geometry, not body details, drives the effect. Output-layer geometry alone therefore reshapes the compute-optimal scaling law of language modeling and brings hyperbolic LMs into the same predictive framework as the Euclidean models that have shaped the field.

Hypergraphs provide a natural representation for high-order relations, but real-world hypergraph data is often difficult to share or collect due to privacy, access, and annotation constraints. Hypergraph generation is therefore useful for constructing surrogate high-order relational data. Existing generators, however, typically build hyperedges through structural heuristics, propagation rules, or projected graph spaces, and thus do not directly model the incidence relation between nodes and hyperedges. Spectral representations offer a principled way to encode global and local relational patterns, but hypergraph spectra are intrinsically non-invertible: a spectral representation does not determine a unique discrete incidence structure. We propose HyperGen, a structure-aware generative framework that models hypergraph generation as continuous dynamics in a phase-enhanced spectral latent space. HyperGen constructs a Hermitian node--hyperedge spectral operator with learnable incidence phases, learns a conditional vector field that transports noise toward target spectral embeddings, and decodes the generated latents into node--hyperedge incidence matrices through a learned spectrum-to-structure mapping with denoising. Experiments on benchmark hypergraphs show that HyperGen more accurately recovers node--hyperedge incidences and better preserves spectral and geometric characteristics than existing baselines.

Hypergraphs capture high-order interactions that ordinary graphs cannot represent, yet generating realistic ones remains challenging because existing methods often rely on fixed structural assumptions or struggle to model coupled dependencies between nodes and hyperedges. In this work, we propose \textsc{Janus}, a latent diffusion framework that decomposes an observed hypergraph into sub-hypergraphs, learns coupled node and hyperedge latent representations with a dual-view $\beta$-VAE, and generates new structures through paired latent diffusion with cross-view conditioning. The method supports both unconstrained and node-set-constrained generation. We also introduce an evaluation protocol covering micro, meso, and macro scales interactions, structural reconstruction, and high-order similarity measures. Across five real-world datasets and nine baselines, \textsc{Janus} achieves the most stable performance across structural scales, with strong gains in meso-scale and high-order fidelity; in the constrained setting, it consistently obtains the best reconstruction results.

Hypergraphs record multi-way interactions among entities. Extracting information from the combinatorial structure underlying observed multi-way interactions is a central task in many real-world problems. Existing methods face several limitations. First, many deep architectures for hypergraphs do not explicitly exploit the potential low-rank structure, which can sacrifice parsimony and interpretability in the learned representations. Second, many low-rank-based methods operate on tensor representations, which typically require hyperlinks to have uniform sizes and thus limit their applicability to general hypergraphs with non-uniform hyperlink sizes. Third, many methods ignore the fact that hyperlinks often arise from heterogeneous mechanisms. For example, medical symptoms may co-occur in the profiles of patients with very different conditions, and such heterogeneity should be incorporated into the learning process. In this work, we develop a general framework for hypergraph representation learning using hyperlink random effects while exploiting the low-rank structure in hypergraphs. The proposed framework accommodates latent heterogeneity in hyperlink formation while preserving entity interaction patterns. We establish identifiability of the model parameters and leverage an expectation-maximization strategy for estimation. The framework allows flexible specifications for the hyperlink random effects; in this paper, we study three choices: categorical, Gaussian mixture, and score-based effects, and develop corresponding estimation algorithms. Through simulation studies, we demonstrate the effectiveness of the proposed method in recovering latent structure and capturing heterogeneous interaction patterns. Empirical studies on real-world hypergraph datasets further illustrate the practical utility of our approach.

Synthetic patient data generation is a promising solution to the dual challenge of data scarcity and privacy constraints in healthcare machine learning. Realistic synthesis of patient-level clinical data requires jointly modeling heterogeneous static covariates, irregularly sampled longitudinal trajectories, and informative observation times — three tightly coupled components in practice yet rarely addressed together. We propose HyperNSDE, a continuous-time generative model that conditions a latent Neural SDE on static patient representations through a hypernetwork, allowing baseline characteristics to shape the full trajectory dynamics rather than only the initial condition, without requiring a trajectory encoder, while stochastic latent dynamics capture realistic variability in generated paths. Observation times are modeled jointly through a latent-state-dependent intensity process. Training on irregular stochastic paths is stabilized via a deterministic—stochastic path decomposition with a non-adversarial signature-kernel objective. Experiments on simulated and real clinical datasets demonstrate consistent improvements over strong baselines across fidelity, privacy, and utility metrics.

Zero-shot machine unlearning aims to remove the influence of specified training instances from a pretrained model when the retain set is unavailable. Instance-wise unlearning further operates at the level of individual instances, where a forget instance may induce instance-specific representation and margin deviations that are entangled with class-relevant local structure. However, without retain instances, separating these deviations from the structure that retraining would preserve is difficult, exposing retain-forget entanglement in representation space. We propose HLMR, a novel zero-shot instance-wise unlearning approach that formulates unlearning as local margin retraction in hyperspherical representation space. HLMR uses hyperspherical geodesic probes to construct a local reference in representation space, preserving the structure supported by this reference while unlearning instance-specific margin excess. Empirical evaluation across multiple dataset--architecture settings shows that HLMR aligns more closely with the retraining oracle in utility and membership-inference risk than state-of-the-art zero-shot baselines.

As foundation models grow in capability, the ability to efficiently and reliably control their behavior becomes critical. Fine-tuning these models can be costly, and while prompting can be practical for controllability, it remains fragile due to models' high sensitivity to exact prompt wording and structure. This brittleness has driven interest in activation steering techniques that offer more stable and predictable control over model behavior. However, existing activation steering methods require per-concept optimization, which makes them ill-suited to deployment scenarios where the concept set is large, evolving, or only specified at request time: each new concept incurs at least minutes of optimization on the target model. We propose HyperTransport, a hypernetwork framework that amortizes this cost by mapping embeddings from a pretrained encoder (CLIP in our instantiation) directly to intervention parameters, trained end-to-end using an optimal transport loss. Once trained, HyperTransport produces each new intervention in a single hypernetwork forward pass, 3600–7000x faster than per-concept fitting. On concepts unseen during training, it matches the strongest per-concept baselines at inducing the target concept. By decoupling concept representation from intervention prediction, HyperTransport combines three capabilities that no existing approach offers as a set: amortized steering for open-ended concept sets, continuous interpretable strength control, and cross-modal conditioning where reference images can directly steer text-based generation. We validate HyperTransport on DMD2 and Nitro-1-PixArt across 167 held-out test concepts via CLIP-based metrics, a VLM-as-a-judge evaluation, and a user study. In pairwise comparisons, both human and VLM judges prefer HyperTransport over prompting ~2x as often.

Real-world image datasets are deeply hierarchical, yet unsupervised hierarchical representation learning remains underdeveloped. Hyperbolic space is geometrically ideal for embedding trees with low distortion due to its exponential volume growth, but existing unsupervised hyperbolic methods either fail to scale beyond a few thousand points or depend heavily on class supervision. We bridge this gap with HyperTree, a fully unsupervised, scalable, and capacity-efficient framework for hierarchical representation learning in hyperbolic space, built on three contributions: (i) a geometrically consistent, closed-form definition of the hyperbolic Lowest Common Ancestor, valid across hyperbolic manifolds of arbitrary curvatures; (ii) a differentiable hierarchical objective based on Dasgupta's cost that is bounded and invariant to dataset scale; and (iii) a sample-complexity guarantee that enables scaling to datasets such as ImageNet-1k and iNaturalist-2018. Across four benchmarks, HyperTree outperforms Euclidean baselines by wide margins on hierarchical dendrogram purity, matches supervised hyperbolic baselines on CIFAR-100, and surpasses them on TinyImageNet at 16 dimensions — a highly capacity-efficient regime where competing methods collapse.

With the rapid growth of surveillance data, Person Re-Identification (ReID) is evolving toward large scale scenarios, where datasets exhibit substantial increases in both size and diversity. Under this trend, the widely adopted dense backbone architectures, such as Transformers, suffer from an inherent limitation: Shared parameters couple all input tokens, forcing the model to learn generic representations that are suboptimal for heterogeneous data sources. Although Mixture-of-Experts (MoE) architectures offer a promising solution for handling diverse data, they encounter a fundamental challenge in ReID. Specifically, person images exhibit high inter-instance similarity at the coarse level, while discriminative cues lie in subtle fine-grained details that are easily overlooked. As a result, existing MoE methods often suffer from severe routing collapse, leading to insufficient expert specialization and limited diverse knowledge acquisition capacity. To address these problems, we propose a novel framework, termed Hyperspherical Mixture-of-Experts for Large Scale Person Re-Identification (HypMoE-ReID). Specifically, motivated by theoretical analysis in hyperspherical space, HypMoE-ReID incorporates two key components: (1) a Token Purification mechanism that suppresses noisy tokens detrimental to expert specialization, and (2) a Routing Diversity Learning strategy that explicitly separates experts on a unit hypersphere to enforce specialization and prevent collapse. Furthermore, to facilitate the investigation in large scale ReID, we construct a large-scale ReID benchmark comprising 809,701 images by unifying 13 public datasets with significant distribution gaps. Furthermore, extensive experiments demonstrate that HypMoE-ReID outperforms state-of-the-art MoE-based methods by at least 11.3\%/10.3\% in Average mAP/R@1, and surpasses strong dense backbone models by 2.2\%/2.0\% in Average mAP/R@1. Our code will be released.

Built on pretrained vision foundation models (VFMs), representation autoencoders (RAEs) have recently emerged as a promising approach for constructing semantically rich latent spaces for image generation. However, their reconstruction quality often remains suboptimal, largely because deep VFM representations do not preserve sufficient fine-grained visual detail. This limitation becomes even more severe after discretization, where missing low-level information is difficult to recover. In fact, we observe that shallow VFM features retain considerably richer local appearance and structural detail, which complements the high-level semantics carried by deep features used in existing RAEs. Motivated by this complementary property, we propose IDEAL, an In-DEpth ALignment framework for discrete representation autoencoding. By jointly aligning quantized tokens with both shallow and deep VFM features, IDEAL enables the resulting discrete visual tokens to preserve both visual fidelity and rich semantics. Extensive experiments demonstrate that IDEAL yields superior reconstruction performance, achieving $\mathbf{0.61}$ rFID on ImageNet and outperforming the previous best method by $\mathbf{0.28}$. When used for autoregressive image generation, IDEAL further produces a gFID of $\mathbf{1.89}$, establishing a new state of the art for autoregressive image generation.

The Linear Representation Hypothesis (LRH) posits that Transformer activations decompose into sparse linear combinations of interpretable directions, with Sparse Autoencoders (SAEs) serving as the predominant method for recovering such decompositions. We propose that idempotency -- the requirement that re-encoding an SAE's output reproduces the same latent representation -- is a desirable property for any faithful dictionary learning method, and demonstrate empirically that most SAEs deviate very far from this ideal. We identify two underlying failure modes: exploding representation norms and unstable active feature sets. Further analysis reveals that even individual SAE latents cannot be reliably reconstructed under re-encoding, and that SAE outputs exhibit significant sensitivity to input scaling. We show that idempotency can be improved via an auxiliary loss term and demonstrate its potential to improve SAE interpretability.

Causal discovery methods commonly assume that all data is independently and identically distributed (i.i.d.) and that there are no unmeasured variables affecting the system. In practice, these assumptions are often violated, leading to inaccurate inference. In this paper, we study how to identify hidden confounding and selection biases from causal mechanism shifts. In particular, we show that structural biases lead to dependent mechanism shifts. That is, by considering for which variables the mechanisms change given data from different environments, we can tell which variables are unbiased, which are subject to hidden confounding, and which are undergoing selection bias. We formalize this into an empirically testable criterion based on mutual information, and show under which conditions it identifies structural biases. To tell which nodes are subject to what kind of bias, we introduce the StruBI algorithm. Experiments on synthetic and real-world data show that StruBI works well in practice, accurately recovering affected variable sets and types of biases, outperforming the state-of-the-art by a wide margin.

Cross-domain time-series generation is a challenging task for diffusion models due to the strong requirement for domain adaptability. While frequency-based Mixture-of-Experts (MoE) methods have shown promise in domain adaptation, their direct integration with diffusion models faces critical issues from intra- and inter-domain. For intra-domain aspect, the absence of frequency transition compliance leads to imbalance pattern learning across denoising steps and degrade model performance. For the inter-domain aspect, the domain-specific transition variation bring difficulties for domain adaptation. To address these challenges, we propose IIDiff, a frequency-based diffusion MoE framework equipped with progressive frequency specializing and dynamic transition matching. The progressive specializing allows experts dynamically adjust frequency receptive fields to capture underlying patterns aligning with the intra-domain transition, enhancing the generation fidelity. The transition matching leverages the frequency noise distribution property to analysis the instantaneous transition process, enabling adaptive routing with domain-specific transition. Extensive experiments on 12 real-world datasets across 4 domains demonstrate our model's superiority. Compared with SOTA baselines, our model achieves an average 27.59\% improvement in KL divergence and 6.00\% improvement in MMD, confirming its robust domain adaptability and ability to generate high-fidelity cross-domain time-series.

Illusory pattern perception is a well-documented human cognitive tendency to infer meaningful relationships in data that is actually random. Such a tendency, often described as “connecting the dots” where none exist, can result in systematic reasoning errors. This paper investigates whether Large Language Models (LLMs) exhibit such perceptual tendencies, which can lead to systematic errors in downstream applications. To our knowledge, this work presents the first systematic study of illusory pattern perception in LLMs, adapting classic psychological paradigms to three tasks with direct empirical comparison to human behaviors. We find that LLMs frequently exhibit stronger illusory pattern perception than humans. In particular, models tend to over-associate frequent positive attributes with majority groups or large organizations, and show increased tendencies to construct causal narratives from ambiguous events. To uncover the mechanism behind these behaviors, we develop a feature interpretability framework based on Sparse Autoencoders (SAEs) to analyze internal representations. Our results reveal that holistic frequency perception and analytic cognitive orientation are linked to the emergence of illusory perceptions. These findings highlight a previously underexplored cognitive-like illusion that may affect the reliability of LLM reasoning. Code available at Anonymous Github.

Embodied navigation aims to build agents that interpret multimodal goals, reason in 3D space, and reach target destinations reliably in the real world. However, progress remains constrained by the lack of scalable, high-fidelity, and physically grounded interactive environments. Although real-world scanned datasets offer visual realism, they are limited by scale. In contrast, synthetic simulators scale more easily but often exhibit large sim-to-real gaps. We introduce Image2Sim, a real-time neural simulation framework that constructs high-quality interactive environments from posed RGB-D image sequences. The central idea is to decouple 3D spatial anchoring from photorealistic observation synthesis. For scene construction, Image2Sim uses a feed-forward feature Gaussian model that lifts posed RGB-D observations into a 3D feature-Gaussian representation in a single pass. For rendering, we propose a Geometry-Aware One-Step Pixel Flow model that transforms sparse and noisy Gaussian projections into high-quality panoramic RGB-D observations. Image2Sim also serves as a fully automated embodied data engine that generates high-fidelity observations, executable actions, and diverse navigation instructions at scale. It converts large collections of videos and images into near 20K interactive scenes and synthesizes more than 10 million navigation training samples. Navigation models trained entirely in these neural environments achieve strong improvements on major benchmarks and transfer effectively to real-world zero-shot settings. These results suggest that scalable neural simulation can serve as a practical training substrate for embodied navigation at scale.

Image matting aims to estimate a continuous alpha matte that separates foreground objects from the background. However, existing deep matting methods still rely heavily on task-specific supervision for matting, including alpha mattes and trimaps, which are costly to annotate and limit their applicability in annotation-scarce scenarios. To address this limitation, we propose FreeMatte, a novel annotation-free framework that learns to predict high-quality alpha mattes without requiring any matting-specific annotations. The proposed framework formulates the construction of matting supervision as a foreground-driven wavefront propagation process. Starting from a coarse localization prior generated by an off-the-shelf segmentation foundation model, it solves an Eikonal equation constructed from the prior and the input image itself to obtain an arrival-time field that encodes how readily each pixel can be reached from the coarse foreground source. The arrival-time field is then translated into confidence-aware sparse supervision, which constrains the matting model to learn alpha prediction from the propagation structure encoded by the field. Extensive experiments across diverse benchmarks show that FreeMatte achieves performance competitive with task-supervised matting baselines under this challenging setting, ranking within the top two on 13 out of 20 evaluated dataset-metric pairs.

Reasoning about the 3D world from multi-view images remains a fundamental challenge for Multimodal Large Language Models (MLLMs). While modern MLLMs handle single-image inputs effectively, they struggle to integrate evidence across viewpoints into a coherent 3D understanding. A growing body of work attempts to close this gap by injecting 3D awareness into MLLMs, either by boosting fine-grained pixel-level cross-view correspondence or by fusing features from 3D geometry foundation models, yet a substantial gap to human reasoning persists. In this work, we revisit human spatial reasoning, which suggests that rather than relying on fine-grained geometry cues, humans roughly identify common objects across views, infer the relative geometry between viewpoints, and assemble a coarse 3D layout of the scene. Inspired by this process, we introduce Imagine3D-LLM, an MLLM that learns to assemble a similar compact 3D representation of the scene and conditions its answer on this representation. Concretely, we append a small set of learnable summary tokens after the image tokens, decode them into a compact 3D Gaussian Splatting representation supervised by a photometric reconstruction loss, and train jointly with the standard next-token prediction objective. Notably, although only the summary tokens receive direct reconstruction supervision, this objective also induces stronger cross-frame correspondence within the LLM's underlying image features, suggesting that learning to reconstruct propagates 3D-aware signals throughout the model. As a result, Imagine3D-LLM consistently outperforms prior approaches across multiple spatial reasoning and 3D understanding benchmarks, suggesting that imagining the scene can be more effective than being told its pixel-wise geometry. Our code and weights will be publicly released.

Uncertainty quantification for individual treatment effects (ITEs) is a daunting challenge in causal inference. Motivated by recent advances in conformal prediction, several works aim to construct distribution-free prediction sets for ITEs with desired coverage under standard assumptions such as strong ignorability and overlap. In this paper, we show that such goals are fundamentally unattainable in the presence of continuous covariates. Specifically, we establish finite-sample and asymptotic impossibility results demonstrating that any distribution-free prediction set achieving desired coverage for ITEs must be trivial, in the sense that it has infinite expected length. Our analysis relies on a connection between ITE inference and the hardness of conditional independence testing, and highlights the intrinsic limitations imposed by the missing data nature of causal inference. These results provide a new perspective on existing methods, clarifying that their apparent success necessarily relies on additional structural assumptions beyond standard causal assumptions.

Retrieval-Augmented Generation (RAG) systems are increasingly deployed in high-stakes domains where users expect outputs to be consistent across semantically equivalent queries. However, existing systems often exhibit significant inconsistencies due to variability in both the retriever and generator, undermining trust and reliability. In this work, we focus on information consistency—the requirement that generated outputs convey the same core content across semantically equivalent inputs. We introduce a principled evaluation framework that decomposes RAG consistency into retriever-level, generator-level, and end-to-end components, enabling systematic identification of inconsistency sources. To improve consistency, we propose Paraphrased Set Group Relative Policy Optimization (PS-GRPO), an RL approach that leverages multiple rollouts across paraphrased set to assign group similarity rewards. We leverage PS-GRPO to achieve Information Consistent RAG (Con-RAG), training the generator to produce consistent outputs across paraphrased queries and remain robust to retrieval-induced variability. Because exact reward computation over paraphrase sets is computationally expensive, we also introduce a scalable approximation method that retains effectiveness while enabling efficient, large-scale training. Empirical evaluations across short-form, multi-hop, and long-form QA benchmarks demonstrate that Con-RAG significantly improves both consistency and accuracy over strong baselines, even in the absence of explicit ground-truth supervision. Our work provides practical solutions for evaluating and building reliable RAG systems for safety-critical deployments.

Classifier-free guidance (CFG) is essential for improving sample quality in visual generation, but it incurs substantial sampling overhead. Existing guidance-free approaches halve this cost by approximating the CFG with a single model inference, but their performance is bounded by the CFG and inherits its limitations. In this paper, we propose self-contrastive guidance (SCG), which contrasts the model's prediction on real and self-generated samples to obtain a direct, model-based signal that pushes generation toward the true data manifold. Integrating SCG into the standard likelihood objective via reparameterization yields Self-Contrastive Alignment for Likelihood Estimation (SCALE), a guidance-free training framework that produces SCG-enhanced predictions in a single forward pass. Across class-conditional and text-to-image generation on diffusion and autoregressive models, SCALE consistently outperforms CFG-mimicking guidance-free baselines and improves further when combined with DPO, demonstrating that the self-contrastive signal yields gains complementary to both CFG and preference-based fine-tuning.

Self-supervised Vision Transformers learn strong image-level representations, but their patch-level representations can be unreliable for dense prediction. We study this mismatch and find that the class token concentrates global semantics on a small number of representative patches, while dense prediction errors often occur on low-attention regions. We provide a local gradient analysis showing that, along the [CLS] attention readout, image-level self-distillation gradients are weighted by [CLS]-to-patch attention mass. To provide masked patches with global semantics without changing the architecture, we propose Cross Distillation (CODI), which mixes a small amount of the teacher class-token distribution into each teacher patch-token target. CODI gives every masked patch an attention-independent semantic correction while preserving the standard self-distillation architecture. Experiments on dense and image-level benchmarks show that CODI improves both dense and image-level performance under aligned pretraining settings.

Latent diffusion models paired with a variational auto-encoder (VAE) tokenizer have shown promising performance for efficient text-driven human motion generation. However, motion latent diffusion suffers from a generation-reconstruction trade-off: simply increasing tokenizer capacity enhances reconstruction but fails to proportionally improve text-conditioned generation quality. We assume this trade-off stems from a spectral mismatch inside the VAE tokenizer: the encoder yields a latent space biased toward high frequencies, and the decoder introduces temporal artifacts. In this paper, we propose the Diffusable Motion Tokenizer (DiMoT) to improve the diffusability of the tokenizer, rather than scaling up the diffusion backbone. To improve text-conditioned optimization, Critic Spectral Adaptation (CSA) leverages a pretrained diffusion model as the critic to suppress high-frequency latent energy. To preserve motion dynamics, the Alpha-Flow Motion Decoder (AMD) performs generative decoding via average-velocity prediction, enabling efficient one-step decoding. Extensive experiments demonstrate that DiMoT achieves state-of-the-art performance across various challenging text-to-motion benchmarks, yielding superior motion fidelity and text-motion alignment. The code will be available upon publication.

Non-convex optimization is central to machine learning and AI, yet unlike its convex counterpart, it still lacks a general and principled framework. In this work, we try to highlight the importance of combining noisy perturbations with over-parametrization mathematically under the classic setting of matrix completion (MC). Matrix completion is an important non-convex recovery problem in which only a subset of a matrix’s entries is observed, and the remaining entries must be recovered using low-rank constraints. This problem is notoriously difficult because observations are limited. However, we show that by injecting $\epsilon$-level perturbations into the nullspace of the observation mask, we can construct noisy matrix sensing surrogates whose solutions remain $\mathcal{O}(\epsilon)$-close to the ground truth with valid restricted isometry property (RIP) constants. The existence of RIP further enables powerful tensor frameworks to be applied with theoretical guarantees. This approach reflects a broader empirical trend in which stochasticity or noise improves the curvature of the optimization landscape, allowing over-parameterized (a.k.a large) models to better separate signal from noise. Assuming each matrix entry is observed independently, we establish quantitative recovery guarantees without explicit requirements on sampling rate or incoherence, and discuss how such assumptions can further strengthen our results.

For LLM agents to be useful beyond simple chat, they must hold sensitive user context such as calendars, credentials, health records, and financial data. We study whether the mere presence of such secrets in a model's context window introduces hidden correlations into the model's benign outputs, allowing reconstruction even when the model correctly refuses direct extraction. We further study whether an adversary can actively engineer prompts that amplify this effect, using the model as a covert carrier to transmit secrets through seemingly innocuous text. In both cases, this limited leakage is exploited using a novel adaptive attack that assumes black-box access to the underlying model. In controlled experiments across eight proprietary models, we find that 2-digit in-context secrets are reconstructed with near-perfect accuracy and 4-digit secrets at 82% exact match, all from outputs the model produces in response to ordinary, non-adversarial requests. More capable models leak more: stronger instruction-following amplifies sensitivity to in-context secrets, making leakage a byproduct of capability rather than a bug to be patched. We show this leakage enables two practical attacks: (1) a trained classifier that infers semantic predicates about user memories (e.g., health conditions, financial events) from routine natural-language outputs, and (2) an RL-trained adversary that extracts full Social Security Numbers from a production-style agent at 100% success.

In this paper, we introduce INEUS, a meshfree iterative neural solver for partial integro-differential equations (PIDEs). The method replaces the explicit evaluation of nonlocal jump integrals with single-jump sampling and reformulates PIDE solving as a sequence of recursive regression problems. Like Physics-Informed Neural Networks (PINNs), INEUS learns global solutions over the entire space-time domain, yet it offers a more efficient treatment of nonlocal terms and avoids the computationally expensive differentiation of full PIDE residuals. These features make INEUS particularly well suited for high-dimensional PDEs and PIDEs. Supported by a contraction-based convergence proof for linear PIDEs, our numerical experiments show that INEUS delivers accurate and scalable solutions for various high-dimensional linear and nonlinear examples.

Computing exact Optimal Transport (OT) distances for large-scale datasets is computationally prohibitive. While entropy-regularized alternatives offer speed, they sacrifice precision and frequently suffer from numerical instability in high-accuracy regimes. To address these limitations, we propose the Inexact Bregman Sparse Newton (IBSN) method, which efficiently solves the exact OT problems. Our approach utilizes a Bregman proximal point framework through a sequence of semi-dual subproblems. By solving these subproblems inexactly, we significantly reduce per-iteration complexity while maintaining a theoretical guarantee of convergence to the true optimal plan. To further accelerate the algorithm, we develop a sparse Newton-type solver for the subproblem and employ a Hessian sparsification strategy that drastically lowers memory and time costs without sacrificing accuracy. We provide rigorous theoretical guarantees for the global convergence of the algorithm. Extensive experiments demonstrate that IBSN consistently outperforms state-of-the-art methods in both computational speed and solution precision.

Flow Matching (FM) provides an efficient formulation for continuous-time generative modeling via straight optimal transport (OT) trajectories. However, during accelerated sampling, these models face a fundamental variance-fidelity dilemma. Under few-step deterministic Ordinary Differential Equation (ODE) integration, the rigid OT trajectories suffer from mean concentration, leading to severe variance collapse and the loss of high-frequency details. Conversely, traditional Stochastic Differential Equations (SDEs) inject blind, exogenous Brownian noise to recover variance, but this isotropic exploration incurs inherent numerical attrition over long integration paths, bottlenecking peak fidelity. To resolve this, we introduce Self-Aligned Drifting (SADrift) to mitigate these challenges without external priors. SADrift is an inference-time framework that augments flow matching inference with self-interacting dynamics, without additional training. By maintaining a momentum history of the sampling trajectory, SADrift extracts the \emph{kinematic residual} to generate a deterministic extrapolative field. This mechanism forces the trajectory to efficiently explore the local tangent space, effectively decoupling exploratory variance into targeted deterministic dispersion and a minimal stochastic buffer. To safely anchor these dispersed trajectories and prevent out-of-distribution divergence, the residual drift is analytically integrated into a micro-time Ornstein-Uhlenbeck (OU) process, providing a mathematically sound, mean-reverting geometric safeguard. Requiring exactly zero additional Neural Function Evaluations (NFE), SADrift empowers a frozen base model to achieve a highly competitive FID of 2.00 at just 20 NFE on ImageNet-256. By successfully bypassing the numerical attrition of the 250-step SDE baseline (FID 2.06), SADrift establishes a new Pareto frontier for variance injection. Beyond image synthesis, we further validate SADrift on fast video generation paradigms (e.g., TurboDiffusion with Wan2.1-1.3B), demonstrating its universal capability to unlock latent generative quality across different modalities and acceleration regimes.

A major goal in computational neuroscience is to understand how animals learn to perform a new behavior from a sequence of actions and rewards. Recent work from Liebana et al. (2025) has argued that animal learning trajectories are inconsistent with learning in shallow (1-layer) networks and are better explained by gradient descent (GD) in a deep neural network (DNN). However, previous work on inferring animal learning rules from behavioral data has focused on either 1-layer networks or fits to group behavior, highlighting the need for methods to fit DNN learning rules to individual animal learning trajectories. To address this gap, we develop a method for inferring single-animal DNN-based learning rules, parametrized by a set of initial weights, a learning rate, and a softmax parameter governing stochasticity in choice behavior. We first prove that GD in a shallow network can be arbitrarily well approximated by GD in a deep network, establishing a theoretical limit on network identifiability. Using this result to refine our inference method, we analyze two behavioral datasets from mice learning a sensory decision-making task. Specifically, we compare three model architectures: 1-layer linear (1L) and 2-layer linear (2L) models considered in Liebana et al., as well as a 2-layer nonlinear network (2NL) with a sigmoidal nonlinearity after the first layer. The 2NL model achieved the best fit for 84 out of 88 mice in both datasets (by AIC). Moreover, simulated learning trajectories from the fitted 2NL model closely matched the timecourse of key behavioral metrics such as bias, stimulus sensitivity, and accuracy. These results show that both depth and nonlinearity are critical for capturing individual learning trajectories. Strikingly, we found that the inferred learning rate and initial layer-1 weights of the 2NL model systematically predicted individual differences in accuracy and task bias, linking network initialization and learning rate to individual differences in learning performance across mice.

Exploring high-order correlations and long-range dependencies in brain networks holds significant value for both neuroscience research and clinical diagnosis. However, previous studies have lacked a unified integration of high-order and long-range dependency information in brain networks, and there is substantial redundancy behind various types of information. These issues limit their effectiveness in the diagnosis of brain diseases. To address this, we propose an Information Bottleneck- Guided Adaptive HyperGraph Transformer (IBAHGT). By incorporating the information bottleneck (IB) principle, this approach enables adaptive learning of high-order correlations and both short- and long-range dependencies within a unified framework for brain network analysis, achieving high-precision brain disease diagnosis. IBAHGT consists of three key components: an information bottleneck- guided adaptive hypergraph convolution, which introduces a novel hypergraph information bottleneck (HIB) principle to adaptively learn hypergraph message- passing weights between nodes and hyperedges, optimizes information flow and captures high-order information in brain networks that is maximally informative and minimally redundant (MIMR). The Transformer encoder captures global information within brain networks through the attention mechanism, specifically modeling short- and long-range dependencies. An information bottleneck-guided node-level adaptive fusion employs the IB principle to learn independent weights for each node, facilitating the fine-grained integration of high-order information and global information to obtain an efficient representation for downstream tasks. Extensive experiments demonstrate that the proposed method outperforms current state-of-the-art methods and can identify biomarkers for clinical applications.

LLMs exhibit uneven performance across programming languages, yet benchmark gaps often reflect narrow task regimes and evaluation-specific confounds. This makes it hard to tell whether poor performance reflects a broader weakness with a given language, a broader difficulty on a given task, or a limitation specific to that task-language combination. We adapt Information Parity, a measure of relative cross-language representational efficiency developed for natural language, to code. We define Code Information Parity (Code IP) on a controlled parallel corpus of 22 algorithms across 20 languages, audited to preserve core semantics while minimizing boilerplate and excluding library shortcuts. The results are regime-specific: Code IP correlates strongly with multilingual benchmark performance on short-horizon generation ($\rho$ from $-0.55$ to $-0.81$), remains negative though moderated on execution prediction ($\rho=-0.429$), and disappears on whole-program synthesis ($\rho=+0.105$). Rather than a universal metric of multilingual coding ability, Code IP isolates a reliable cross-language signal on short-horizon generation and delineates where that signal no longer separates cleanly from broader task and evaluation effects.

Classical initialization theory analyzes networks through second-order observables such as variance, correlation, and Jacobian conditioning that are natural when weights are symmetric around zero. Identity-plus-noise initialization, which initializes the network to behave locally like a residual block, sits outside this picture: variance can be preserved while coordinate-wise signal identity is destroyed, making such second-order quantities indirect indicators of trainability. We identify the terminal sign-flip rate as the natural forward-pass observable for this regime, and derive its critical scale $\sigma \sim L^{-1/2}$ from a dimensionless margin--sensitivity ratio, rather than postulating it. The retention curve emerges as a first-passage limit of a signed-margin process, yielding a calibration procedure that targets a chosen sign-flip rate at any depth. Experiments on synthetic and real benchmarks suggest that this calibrated initial rate behaves as a training-time parameter rather than an inert calibration detail.

Many modern learning criteria, such as multicalibration, CVaR, and distributionally robust objectives, are set-input and optimization-valued: their empirical values depend on the whole evaluation batch and on an auxiliary optimizer selected after observing that batch. This optimization mismatch falls outside the standard information-theoretic generalization analysis for averages of fixed per-example losses. We provide a unified supersample-based analysis for this class of objectives. Under a selector-gap concentration condition, we show that the optimized validation--training gap is controlled by the conditional mutual information between the learned supersample predictions and the selector, without an explicit complexity penalty for the auxiliary optimizer class. We verify the concentration condition through stability and convex L_2-Lipschitz certificates, obtaining concrete bounds for multicalibration, CVaR, and Cressie--Read DRO, including a single-index refinement under stability.

Quantization is a fundamental tool used to compress datasets, neural network weights, and memory usage in a range of computational tasks. Many downstream applications of vector quantization perform inner products with arbitrary inputs. This motivates the study of inner product aware quantization schemes that approximately preserve inner products with unseen vectors -- in contrast to simply minimizing the mean-squared error. In this work, we formulate objectives that capture natural desiderata and develop adaptive and unbiased quantization methods that approximately preserve inner products with worst-case and average case inputs. An analysis of these objectives shows a tight connection with the well-studied notion of Adaptive Stochastic Quantization (ASQ). We develop provably fast exact and approximate algorithms for our objectives. Our theoretical results inspire efficient practical algorithms that perform well across a variety of workload distributions. They also lead to practical algorithms for standard ASQ which are 2-10$\times$ faster than prior state-of-the-art methods while maintaining quality. These theoretical and empirical results contribute towards making adaptive quantization techniques more efficient and tractable in practical settings.

Visual insect specimen understanding aims to identify biological taxa and characterize morphology-relevant structures from specimen images. Existing insect benchmarks have advanced this problem through large-scale pretraining and multimodal learning, but remain centered on taxonomic recognition with strong auxiliary modalities. We introduce \textbf{InSpect}, a curated natural history collection dataset containing 48,184 digitized insect specimen images with aligned crops, hierarchical taxonomy, label-derived structured metadata, and 1,859 images with fine-grained anatomical part masks. We thus establish two benchmark settings. The first is \textbf{open taxonomic recognition}, which evaluates zero-shot, fine-tuned, and unseen-taxon recognition using cropped insect images, label-derived metadata, or their combination. The second is \textbf{fine-grained anatomical part segmentation}, which evaluates supervised and text-guided segmentation of insect structures such as antennae, legs, wings, and body regions. Experiments show that open taxonomic recognition remains challenging despite strong closed-set performance, and that specimen metadata provides useful but unstable contextual cues. For anatomical segmentation, closed-set segmentation models remain weak on thin structures, and zero-shot open-vocabulary models often fail to distinguish insect parts from the body or background. Together, these results show that InSpect enables systematic evaluation of insect specimen understanding across cropped visual morphology, label-derived context, and fine-grained anatomical annotations.

Direct Preference Optimization (DPO) and its variants have become the standard for aligning Large Language Models (LLMs). However, we identify two fundamental limitations. First, the optimized policy lacks invariance since it varies with modeling choices such as scalarization function or reference policy, whereas an optimal policy should remain invariant. Second, most existing methods yield theoretically suboptimal policies by not fully exploiting the comparative information in pairwise preference data, thus missing an opportunity for self-reflection through comparing and contrasting responses. To address both limitations, we propose Intrinsic Self-reflective Preference Optimization (InSPO), which derives a globally optimal policy conditioned on both context and alternative response, explicitly formalizing self-reflection. We prove this formulation surpasses standard DPO and RLHF targets while guaranteeing invariance. InSPO serves as a plug-and-play enhancement for DPO-family algorithms, decoupling alignment from modeling constraints without architectural changes. Using privileged information learning, InSPO requires no alternative response at inference since the self-reflective mechanism is distilled during training, incurring zero overhead. Experiments show InSPO consistently improves win rates and length-controlled metrics across DPO variants, yielding more robust and human-aligned LLMs. Code is available here.

Modality following is the ability to selectively leverage multimodal contexts based on user instructions. It is fundamental to the safety and reliability of multimodal large language models (MLLMs) in real-world deployments. However, the internal mechanisms governing this decision-making process remain largely under-explored. In this work, we investigate the mechanism underlying modality following through an information flow perspective. Our findings reveal that instruction tokens serve as structural anchor for modality arbitration: Shallow attention layers perform undifferentiated information transfer, routing multimodal cues to instruction tokens as a latent buffer; in contrast, deep attention layers selectively strengthen the instruction-compliant subspace and resolve modality arbitration according to the instruction-specified intent, with a sparse subset of attention heads driving this process. Targeted attention-head interventions further validate the functional specificity of these heads: blocking only $5\%$ of the identified heads substantially degrades modality following while preserving general visual and language capabilities, whereas targeted amplification can restore failed modality-following samples by up to approximately $60\%$. Together, this work provides a mechanistic account of modality following and informs future efforts to improve how MLLMs integrate and utilize multimodal evidence under user instructions. \footnote{All data and code will be released upon acceptance.}

We study supervised classification problems with missing feature values. Existing approaches often decouple imputation from classification, producing imputations that may be plausible but uninformative for prediction. Instead, we propose the Integrated Imputation and Classification Network (IICN), which jointly trains an imputer and an ${(n{+}1)}$-classdiscriminator adversarially with a single class supervised classification objective, where the discriminator learns to distinguish among the $n$ true classes and an additional ``imputed" class. We prove that at the global optimum, the imputer and discriminator together implement marginalization over missing coordinates and yield a Bayes-optimal classifier. We evaluate IICN on FashionMNIST, CIFAR-10, and tabular datasets with naturally occurring missingness. IICN outperforms classical impute-then-classify pipelines and recent generative baselines, showing strong robustness and accuracy in challenging settings.

Multi-Head Attention (MHA) is the core computational primitive underlying modern Large Language Models (LLMs). However, within a single attention layer, MHA has an intrinsic linear scaling limitation: $H$ attention heads produce exactly $H$ independent attention matrices, with no communication between heads during attention computation. While deep Transformers can compose information across layers, this places much of the burden of building and combining intermediate relations on depth, potentially requiring additional layers to expose interactions useful for downstream computation. Our focus is complementary: increasing the compositional bandwidth available within each attention layer. To this end, we propose Interleaved Head Attention (IHA), which enables within-layer cross-head mixing by constructing $P$ pseudo-heads per head, typically with $P=H$, where each pseudo query, key, and value is a learned linear combination of all $H$ original queries, keys, and values, respectively. Interactions between pseudo-query and pseudo-key heads induce up to $P^2$ attention patterns per head with modest parameter overhead $\mathcal{O}(H^2P)$. We provide theory showing improved parameter efficiency on Polynomial Filters, where IHA uses $\Theta(\sqrt{k}n^2)$ parameters versus $\Theta(kn^2)$ for MHA, and on the order-sensitive CPM-3 task, where IHA uses $\lceil\sqrt{N_{max}}\rceil$ heads versus $N_{max}$ for MHA. On real-world benchmarks, IHA improves Multi-Key retrieval on RULER by 10–20% at 4k–16k context lengths and, after OpenThoughts reasoning fine-tuning, improves GSM8K by 5.8% and MATH-500 by 2.8% under majority voting over full attention, with modest throughput overhead.

Existing benchmarks for multimodal agentic search evaluate multimodal search and visual browsing, but visual evidence is either confined to the input or treated as an answer endpoint rather than part of an interleaved search trajectory. We introduce \textbf{InterLV-Search}, a benchmark for Interleaved Language-Vision Agentic Search, in which textual and visual evidence is repeatedly used to condition later search. It contains 2,061 examples across three levels: active visual evidence seeking, controlled offline interleaved multimodal search, and open-web interleaved multimodal search. Beyond existing benchmarks, it also includes multimodal multi-branch samples that involve comparison between multiple entities during the evidence search. We construct Level 1 and Level 2 with automated pipelines and Level 3 with a machine-led, human-supervised open-web pipeline. We further provide InterLV-Agent for standardized tool use, trajectory logging, and evaluation. Experiments on proprietary and open-source multimodal agents show that current systems remain far from solving interleaved multimodal search, with the best model below 50% overall accuracy, highlighting challenges in visual evidence seeking, search control, and multimodal evidence integration. We release the benchmark data and evaluation code at \url{https://anonymous.4open.science/r/InterLV-Search-F423}

Existing research infrastructure is fundamentally document-centric, providing citation links between papers but lacking explicit representations of methodological evolution. In particular, it does not capture the structured relationships that explain how and why research methods emerge, adapt, and build upon one another. With the rise of AI-driven research agents as a new class of consumers of scientific knowledge, this limitation becomes increasingly consequential, as such agents cannot reliably reconstruct method evolution topologies from unstructured text. We introduce Intern-Atlas, a methodological evolution graph that automatically identifies method-level entities, infers lineage relationships among methodologies, and captures the bottlenecks that drive transitions between successive innovations. Built from $1{,}030{,}314$ papers spanning AI conferences, journals, and arXiv preprints, the resulting graph comprises $9{,}410{,}201$ semantically typed edges, forming a queryable causal network of methodological development. To operationalize this structure, we further propose a self-guided temporal tree search algorithm for constructing evolution chains that trace the progression of methods over time. We evaluate the quality of the resulting graph against expert-curated ground-truth evolution chains and observe strong alignment. In addition, we demonstrate that Intern-Atlas enables downstream applications in idea evaluation and automated idea generation. We position methodological evolution graphs as a foundational data layer for the emerging field of automated scientific discovery.

Inferring latent interaction structures from observed dynamics is a fundamental inverse problem in many-body interacting systems. Most neural approaches rely on black-box surrogates over trainable graphs, achieving accuracy at the expense of mechanistic interpretability. Symbolic regression offers explicit dynamical equations and stronger inductive biases, but typically assumes known topology and a fixed function library. We propose $\textbf{COSINE}$ ($\textbf{C}$o-$\textbf{O}$ptimization of $\textbf{S}$ymbolic $\textbf{I}$nteractions and $\textbf{N}$etwork $\textbf{E}$dges), a differentiable framework that jointly discovers interaction graphs and sparse symbolic dynamics. To overcome the limitations of fixed symbolic libraries, COSINE further incorporates an outer-loop large language model that adaptively prunes and expands the hypothesis space using feedback from the inner optimization loop. Experiments on synthetic systems and large-scale real-world epidemic data demonstrate robust structural recovery and compact, mechanism-aligned dynamical expressions. Code: https://anonymous.4open.science/r/COSINE-6D43.

We introduce a simple post-training method that makes transformer attention sparse without sacrificing performance. Applying a flexible sparsity regularisation under a constrained-loss objective, we show on models up to 7B parameters that it is possible to retain the original pretraining loss while reducing attention connectivity to $\approx 0.4$% of its edges. Unlike sparse-attention methods designed for computational efficiency, our approach leverages sparsity as a structural prior: it preserves capability while exposing a more organized and interpretable connectivity pattern. We find that this local sparsity cascades into global circuit simplification: task-specific circuits involve far fewer components (attention heads and MLPs) with up to 100× fewer edges connecting them. Additionally, using cross-layer transcoders, we show that sparse attention substantially simplifies attention attribution, enabling a unified view of feature-based and circuit-based perspectives. These results demonstrate that transformer attention can be made orders of magnitude sparser, suggesting that much of its computation is redundant and that sparsity may serve as a guiding principle for more structured and interpretable models.

Operator-valued concentration inequalities are foundational to the analysis of modern high-dimensional statistics and randomized algorithms. However, standard oracle bounds are frequently limited in practice: they require explicit a priori knowledge of the true variance, and often explicitly scale with the ambient dimension, rendering them vacuous for infinite-dimensional or heavily structured operators. Motivated by these challenges, we establish the first empirical Bennett and Bernstein inequalities for sums of independent, bounded, compact self-adjoint operators. Our fully data-driven bounds replace the unknown variance with an empirical estimate and rely strictly on the intrinsic dimension rather than the ambient dimension. This structural shift yields computable, dimension-free guarantees that are strictly sharper for non-isotropic random matrices and seamlessly extend to infinite-dimensional Hilbert spaces. We demonstrate that our empirical bounds achieve asymptotic sharpness with the best known oracle rates. Finally, as an independent byproduct, we derive novel empirical concentration guarantees for the intrinsic dimension itself.

Recent work due to Goel et al.~gave the first efficient algorithms for learning with distribution shift in the challenging PQ framework, where a learner receives labeled training examples and unlabeled test examples and must make correct predictions on the test set but is allowed to abstain from predicting on out-of-distribution points. Their results rely on ${\cal L}_2$ sandwiching approximations, a strong requirement that leads to poor bounds for several basic function classes such as DNF formulas. Here we show that the weaker notion of ${\cal L}_1$ sandwiching suffices for efficient PQ learning. As a consequence, we obtain the first quasipolynomial-time PQ learning algorithm for DNFs under the uniform distribution and essentially match the guarantees known for ordinary PAC learning. More broadly, our bounds provide exponential improvements for several classes including constant depth circuits and constant degree polynomial threshold functions. Our main technical ingredient is *Iterative Chow Filtering*, a new procedure that uses low-degree Chow parameters to identify and remove test points incompatible with the training distribution.

Gradient-based attribution methods for large language models (LLMs) have so far been limited to explaining scalar outputs - the logit of a single target token. Yet LLM predictions are inherently distributional, and many practically important questions concern the full predictive distribution: which tokens make the model uncertain? Which drive a broad range of plausible continuations? We propose **Jacobian Scopes**, a unified framework that fills this gap by projecting the input-to-output Jacobian onto different directions in output space via a single vector-Jacobian product, requiring only one backward pass. This yields three complementary methods: **Semantic Scope** attributes a specific target logit; **Fisher Scope**, grounded in information geometry, identifies tokens that most alter the overall shape of the predicted distribution; and **Temperature Scope** traces which tokens govern the model's predictive confidence. Fisher and Temperature Scopes are, to our knowledge, the first attribution methods to natively target distributional rather than pointwise features of LLM predictions. Through case studies spanning instruction following, translation, and in-context time-series forecasting, Jacobian Scopes reveal implicit political biases, uncover word- and phrase-level translation strategies, and illuminate the nearest-neighbor pattern-matching mechanisms underlying LLM forecasting. Quantitative evaluation on LAMBADA and IWSLT2017 across six leading LLMs (LLaMA-3.2, Qwen2.5, Gemma-3) confirms that Jacobian Scopes consistently match or outperform Input $\times$ Gradient and Integrated Gradients, at a fraction of the latter's computational cost.

Inherent temporal heterogeneity, such as varying sampling densities and periodic structures, has posed substantial challenges in zero-shot generalization for Time Series Foundation Models (TSFMs). Existing TSFMs predominantly rely on massive parameterization to absorb such heterogeneity, as their static tokenization and positional encoding schemes entangle diverse temporal patterns into a fixed representation space, encouraging memorization rather than adaptation. To address this limitation, we propose Kairos, a flexible and parameter-efficient TSFM dedicated to forecasting tasks, which decouples temporal heterogeneity from model capacity through a novel tokenization perspective. Kairos introduces a dynamic patching tokenizer and a mixture-of-size encoding that adapt observational granularity to local information density, enabling fine-grained temporal abstraction without increasing model width or depth. In addition, we design a multi-granularity positional embedding based on dynamic rotary encodings, which conditions on instance-level spectral features and temporal structure induced by dynamic patching tokenization, allowing robust modeling of diverse temporal dependencies. Trained on a novel Predictability-Stratified Time-Series (PreSTS) corpus, Kairos achieves superior zero-shot performance with substantially fewer parameters on two mainstream benchmarks, GIFT-Eval and Time-Series-Library.

On-Policy Distillation (OPD) has emerged as a dominant post-training paradigm for large language models, especially for reasoning domains. However, OPD remains unstable in practice due to the high gradient variance of its single-sample Monte Carlo estimator, and recipes for stable training are still immature. We propose vOPD (On-Policy Distillation with a control variate baseline), which casts OPD as policy-gradient RL and stabilizes it by introducing a control variate baseline—canonically a value function—from the RL literature. We show that the OPD value function admits a closed form as the per-token negative reverse KL divergence between the student and the teacher, available directly from the already-computed forward pass with no additional critic or inference. Existing stabilization methods either compute the full token-level reverse KL over the entire vocabulary, adding significant overhead, or restrict it to a top-k support, biasing the objective. vOPD instead preserves the lightweight single-sample estimator, subtracting the value function as a detached baseline to keep the gradient unbiased while reducing variance. Furthermore, we show that a top-k approximation of the baseline further lowers cost without compromising performance. Across mathematical and scientific reasoning benchmarks, vOPD consistently outperforms vanilla OPD and matches the most expensive full-vocabulary baseline, offering an efficient stabilization of On-Policy Distillation through principled RL variance reduction.

Large language model (LLM) unlearning aims to selectively remove specific knowledge while preserving model utility. Current benchmarks such as TOFU, WMDP, and MUSE evaluate unlearning on semantically disjoint forget and retain sets, making the task artificially easy. We formalize this gap through a taxonomy of knowledge entanglement at three levels: entity, concept, and skill. Based on this taxonomy, we construct KNOT, a benchmark with approximately 5,300 forget, 9,800 retain, and 5,000 boundary QA pairs, each annotated with a quantitative entanglement score. Evaluation of seven methods on KNOT reveals that (a) all methods degrade severely under high entanglement, (b) the coreset effect reported on standard benchmarks vanishes, and (c) existing metrics miss critical boundary behavior. We propose Boundary Precision and Boundary Recall to fill this gap. Cross-benchmark comparison confirms that KNOT's entanglement, rather than surface difficulty, drives the observed degradation.

Post-training quantization (PTQ) of diffusion transformers (DiTs) to W4A4 severely degrades output quality, because activations entering each linear layer contain outliers that 4-bit formats cannot represent. The standard fix applies an invertible linear transform to the activations and its inverse to the weights before quantizing both. Normalization layers between blocks force this transform to run online at every denoising step, making its inference computation cost the binding design constraint. Existing options trade quantization quality for inference cost: per-channel scaling (SmoothQuant) is computationally cheap but impacts the magnitude of the channels, which can harm quantization accuracy; fixed Hadamard transforms yield better quantization accuracy but require large block sizes that incur a high online cost; learned full-$d$ invertible transforms calibrate best but entail an prohibitive dense $d \times d$ matrix multiplication (GEMM) per layer per step. We propose KroQuant, a PTQ method that applies a learned Kronecker-structured invertible transform to each 32-element block of the activation, storing less than half the parameters of per-channel scaling. The block-local structure runs as small tensor-core GEMMs, and on an MI350 GPU the KroQuant quantizer kernel is up to $14$\% faster than the SmoothQuant kernel. Offline LoRaQ weight calibration then absorbs the residual per-weight quantization error. On PixArt-$\Sigma$, SANA, and FLUX.1-schnell at W4A4 (MXFP4e2), KroQuant produces outputs closer to the FP reference than SVDQuant and LoRaQ on MJHQ-30K and SDCI, while preserving or improving image quality.

What limits KV-cache compression at extreme bit-rates? We argue it is not the choice of compression scheme but how its budget is allocated across attention heads. Existing methods apply rank and bit-width uniformly, ignoring that each head has a different optimal mix of rank truncation and quantization. We show that co-optimizing rank and bit-width per head — using only standard low-rank projection and scalar quantization — dominates uniform allocation, with the largest gains at low bit-rates. Our method, KV-COBRA (Co-Optimized Bit-Rank Allocation), formalizes this as a resource-allocation problem: it balances rank-truncation loss against quantization loss within each head, then redistributes budget across heads to minimize total distortion. A fused Hadamard rotation equalizes per-channel variance, and reordering the SVD basis by attention-KL importance makes the solver query-aware. The same allocator extends to joint $K{+}V$ compression. On perplexity, zero-shot, and long-context benchmarks from $0.5$ to $4$ bits per dimension (bpd), KV-COBRA shows the smallest accuracy degradation among evaluated methods at low bpd, with no per-token overhead.

We present LAION-BVD (LAION - Big Video Dataset), a large-scale open video dataset for multimodal learning, containing 1.3B platform-specific video URLs collected from CommonCrawl. From these, we download 80M videos with a total duration of 10 million hours. The dataset is designed for multimodal pre-training across video, audio, and image modalities. Using content-aware scene detection we extract clips for which we synthetically generate video and audio captions. Models trained on these data achieve competitive performance on standard video-text and audio-text benchmarks, with consistent improvements as training or model scale increases. Additionally, we explore video frames as a new source of image-text data by extracting scene-changing frames. These frames exhibit a visual distribution distinct from standard web image corpora, and models trained on this dataset achieve strong image-text retrieval performance. Overall, LAION-BVD significantly expands open access to multimodal videos at unprecedented scale and will be released to the research community for multimodal research

Language-conditioned goal navigation (LGN) requires agents to locate user-specified targets without step-by-step guidance. However, existing benchmarks largely focus on category-level goals or rely on instance descriptions generated by vision-language models (VLMs), which often contain ambiguities and semantic errors, limiting systematic and reliable evaluation. We introduce HieraNav, an open-vocabulary LGN task with goals specified at four hierarchical semantic levels: scene, room, region, and instance. To this end, we present Language as a Map (LangMap), to our knowledge the first real-world 3D indoor navigation benchmark with human-verified semantic annotations to support tasks across all four goal levels. LangMap provides region labels and discriminative region and instance descriptions covering 414 object categories, produced through a rigorous contrastive annotation protocol comparing same-scene regions and instances, and contains over 18K tasks. Each target is paired with concise and detailed descriptions, enabling evaluation across instruction styles. Quantitative and qualitative analyses validate our annotation quality; notably, our instance descriptions outperform GOAT-Bench annotations by 23 percentage points in text-to-view matching. We further introduce PlaNaVid, a strong RGB-only baseline that combines Bounded Diverse Memory (BDM) with high-level planning to prime a reactive policy for multi-goal navigation. PlaNaVid achieves top-tier success rates without depth, 3D scene representations, or object masks. Further analysis shows that memory and richer context boost performance, while long-tailed categories, small objects, distant targets, and multi-goal completion remain open challenges.

The emergence of large language models (LMs) as versatile problem-solving systems marks a paradigm shift comparable to the advent of digital computers. Traditional computer science provides theoretical foundations for understanding computational processes through complexity theory, algorithms, and formal methods, this position paper argues that language models require their own distinct theoretical framework, termed as Computer Science of Language Models (CSLM). We identify seven fundamental differences between traditional computing and language model: 1) the generating-solving-verifying triad versus the solving-verifying dichotomy, 2) token-based rather than operation-based computation, 3) probabilistic reliability with hallucinations rather than deterministic correctness, 4) native multimodal understanding beyond pixel-level processing, 5) general-purpose capabilities versus problem-specific algorithms, 6) essential two-stage training-inference processes, and 7) inherent knowledge. We argue these differences necessitate new theoretical frameworks, complexity measures, and evaluation methodologies. Then, we define the token complexity as the core concept in CSLM and discuss connections to existing research in scaling laws, memory systems, tool use, and fine-tuning, while identifying critical open questions about LM-complete problems, reliability bounds, emergence phenomena, and multimodal complexity. Drawing parallels to complexity theory's 50-year development, we present a 5-year research roadmap (2026-2030) to build the mature CSLM. CSLM promises to transform LM development from empirical scaling to principled engineering grounded in rigorous theory and strengthen our understanding about both the problems, LMs and their relationships.

Transformers update token representations through multi-head attention and residual connections as $X \leftarrow X + \sum_{i} P^{(i)}XW_{V_i}W_{o_i}$, where $P^{(i)}$ is the softmax attention matrix in head $i$. We propose replacing a subset of $P^{(i)}$'s with the Laplacian $I - P^{(i)}$, giving $X \leftarrow X + \sum_{i \in \mathcal{A}} P^{(i)}XW_{V_i}W_{o_i} + \sum_{i \in \mathcal{L}} (I - P^{(i)})XW_{V_i}W_{o_i}$. Our proposal has two motivations. First, it allows attention heads to update the mean of token representations, while Laplacian heads can directly control within-sequence variance. Second, if tokens are viewed as nodes in a graph with edge weights \(P^{(i)}\), then \(I - P^{(i)}\) is the corresponding graph Laplacian, and the update can be interpreted as one step of heat diffusion on the graph. We show that this simple modification improves performance across supervised learning, language modeling, and self-supervised learning tasks. To investigate why, we examine the token representations learned with and without Laplacian heads. In supervised learning, Laplacian heads collapse token representations within the same sequence and align the sequence means with the geometry of Neural Collapse. In language modeling, they increase the separability of token representations that share the same next-token prediction. In self-supervised learning, they produce token representations whose principal components are better suited for segmentation. Across modalities, they also lead to faster-decaying spectra, indicating stronger token smoothing. Overall, our findings challenge the prevailing view that token oversmoothing is inherently harmful, showing instead that certain forms of smoothing can be beneficial.

Robotic foundation models require reasoning over complex visual scenes to execute adaptive actions in dynamic environments. While recent studies on latent-reasoning Vision-Language-Action (VLA) models have demonstrated the capability to capture fine-grained physical dynamics, they remain predominantly confined to static imitation learning, severely limiting their adaptability and generalization. In this paper, we present LaST-R1, a novel reinforcement learning (RL) post-training framework designed to effectively harness latent reasoning-before-acting policies. Specifically, we propose Latent-to-Action Policy Optimization (LAPO), a core RL algorithm that jointly optimizes the latent reasoning process and the action generation. By explicitly embedding latent Chain-of-Thought (CoT) reasoning directly within the RL optimization loop, LAPO stimulates profound physical world modeling, which in turn drives robust execution in interactive environments. Furthermore, an adaptive latent CoT mechanism is introduced, allowing the policy to dynamically modulate its reasoning horizon based on diverse environment states. Experiments show that LaST-R1 achieves a near-perfect 99.9% average success rate on the LIBERO benchmark with only one-shot supervised warm-up, significantly improving convergence speed and performance over prior state-of-the-art (SOTA) methods. In real-world deployments, LaST-R1 yields up to a 22.5% average improvement over SOTA supervised fine-tuning approach across four complex tasks, including both single-arm and dual-arm settings. Finally, LaST-R1 demonstrates strong generalization across simulated and real-world environments.

Single-step retrieval-augmented generation (RAG) provides an efficient way to incorporate external information for simple question answering tasks but struggles with complex questions. Agentic RAG extends this paradigm by replacing single-step retrieval with a multi-step process, in which the large language model (LLM) acts as a search agent that generates intermediate thoughts and subqueries to iteratively interact with the retrieval system. This iterative process incurs substantial latency due to the autoregressive generation of lengthy thoughts and subqueries. To address this limitation, we propose LatentRAG, a novel framework that shifts both reasoning and retrieval from discrete language space to continuous latent space. Unlike existing explicit methods that generate natural language thoughts or subqueries token-by-token, LatentRAG produces latent tokens for thoughts and subqueries directly from the hidden states in a single forward pass. We align LLMs with dense retrieval models in the latent space, enabling retrieval over latent subquery tokens and supporting end-to-end joint optimization. To improve transparency and encourage semantically meaningful latent representations, we incorporate a parallel latent decoding mechanism that translates latent tokens back into natural language. Extensive experiments on seven benchmark datasets show that LatentRAG achieves performance comparable to explicit agentic RAG methods while reducing inference latency by approximately 90%, substantially narrowing the latency gap with traditional single-step RAG.

Safety-aligned language models are trained to refuse harmful requests, yet refusal behavior can be suppressed by steering their internal representations. Existing methods do so by ablating a refusal direction from model activations, aiming to remove refusal from the model’s residual stream. Despite their empirical success, these methods lack a principled account of the latent-space transformation they induce and why it suppresses refusal. In this work, we recast refusal suppression as a latent-space evasion attack against linear probes trained to separate refused from answered prompts. Under this view, prior work’s difference-in-means direction naturally defines such a probe, and its ablation is exactly a projection onto its decision boundary, i.e., a minimum-confidence evasion attack. This perspective not only explains the empirical success of prior work but also admits a key limitation: evasion stops at the decision boundary, motivating the need to push representations further into the compliant region, i.e., where the model answers. We leverage this by proposing a Controlled Latent-space Evasion attack that projects representations past the boundary with an optimized confidence. We achieve state-of-the-art attack success rate across 15 instruction-tuned, multimodal, and reasoning models, outperforming existing refusal-ablation baselines and specialized jailbreak attacks.

Despite the remarkable progress of evolutionary LLM frameworks like AlphaEvolve in program and algorithm discovery, they consistently struggle with tasks requiring deep algorithmic reasoning and long-horizon planning, such as the Abstraction and Reasoning Corpus (ARC-AGI). A fundamental limitation of current evolutionary code generation is its reliance on unguided stochasticity to derive child programs from parents, lacking a systematic mechanism for knowledge accumulation, whereas human problem-solving thrives on the continuous consolidation and reuse of high-level conceptual abstractions. To bridge this gap, we propose the \textbf{Library-driven Evolutionary Abstraction Paradigm (LEAP)}, a cognitive-inspired framework that enables LLMs to autonomously construct and evolve a globally shared library of reusable algorithmic primitives. The evolution of this library is governed by two rigorous theoretical principles: First, the generation of new primitives is driven by a Minimum Description Length (MDL) objective, ensuring the system only assimilates concepts that genuinely compress the problem space. Second, the evolutionary survival and selection of these primitives are governed by a principled exploration-exploitation mechanism, dynamically regulating the library's life-cycle to strike an optimal balance between exploiting proven cognitive operators and exploring novel hypotheses while strictly bounding context entropy. Extensive experiments demonstrate that LEAP significantly outperforms existing LLM evolutionary baselines on the ARC-AGI benchmark and math optimizations problems, exhibiting robust cross-task generalization, highly efficient context utilization, and autonomous concept discovery capabilities.

Traffic flow forecasting plays a pivotal role in intelligent transportation systems. However, most existing methods adopt a black-box learning paradigm, resulting in a lack of interpretability in their decision-making processes. Existing explainable methods mostly follow the feature attribution paradigm, leaving explanations at the level of static or discrete feature evidence and making it difficult to reveal the continuous evolution mechanism of spatio-temporal dependencies during dynamic traffic propagation. Meanwhile, prior methods often decouple spatial and temporal explanations, treating spatial structures and temporal dynamics independently and thus failing to capture the intrinsic coupling between spatial dependencies and temporal contexts in model decision-making. To this end, we propose STGIB, a new spatio-temporal graph information bottleneck theory for characterizing continuously evolving explanations in traffic forecasting. STGIB redefines traffic explanation as a time-indexed spatio-temporal explanation trajectory, and jointly learns dynamic key structures and their temporal representations under predictive sufficiency and information compression constraints. Furthermore, we derive tractable recursive variational bounds to make the theoretical objective optimizable, and instantiate a new explainable traffic flow forecasting model via spatio-temporal explanation trajectory generation. Experiments on real-world traffic datasets demonstrate that STGIB maintains competitive predictive performance while generating more faithful and spatio-temporally consistent explanations.

Neural Ratio Estimation (NRE) is a popular method for performing simulation-based posterior estimation given a known prior and samples from a joint distribution. For such a task, it can be crucial for the posterior estimation task to be \emph{statistically efficient}, i.e. to require as few simulations as possible to obtain an accurate posterior estimate. However, currently, little is known about its statistical efficiency. In this work, we bridge this gap by performing an asymptotic statistical analysis of NRE. We show that even when the target distribution is a simple product of Gaussians, the accuracy of NRE can degrade exponentially badly with the dimension of the data distribution. Our results show that NRE can be significantly less efficient than other conditional density methods like Maximum Likelihood Estimation, which are known to enjoy favorable statistical properties.

Does a neuron’s shape predict whom it connects to? Peters’ rule, a canonical geo- metric principle in connectomics, predicts connectivity from spatial arbor overlap, but true wiring also depends on molecular and morphological specificity. We ask how far neuron skeletons, with identity and graph features withheld, predict directed connections in the FlyWire Drosophila connectome. Because most non- connected pairs are far apart, headline metrics are dominated by easy negatives that geometry can reject. We therefore evaluate controlled contrasts: plausible non-connections whose arbors are nearby, overlapping, or cell-type matched, but that still do not synapse. Across a controlled model ladder, from arbor overlap and hand-crafted morphology baselines to Connectoformer (a bidirectional Trans- former with pooled and pairwise cross-attention scoring), performance improves when the model gains the corresponding signal: global shape, local branch com- patibility, and tree context. Morphology and arbor co-occupancy are complemen- tary: morphology discriminates similar-looking unconnected pairs but can score pairs whose arbors never touch; arbor co-occupancy does the inverse. By suppress- ing essentially-zero-overlap pairs without boosting high-overlap pairs, a fixed one- sided arbor prior performs well in both regimes. The encoder also transfers across Drosophila individuals: zero-shot evaluation on the male CNS connectome beats arbor overlap on hard-negative discrimination. Stratified evaluation reveals a clean division of labor: geometry rules out pairs that cannot physically connect, and morphology picks the actual partners from what remains.

Learning flexible motor primitives is a hallmark of skilled motor control. Recent neuroscience theory proposes that motor primitives may be implemented as low-rank perturbations of a shared recurrent network, but leaves open how such a system is learned. We translate this principle into a novel architecture for learning motor skills end-to-end: a shared recurrent core modulated by a bank of residual adapters, each selected by a discrete latent code. Trained on closed-loop biomechanical control, the adapters develop emergent low-rank perturbations of the recurrent dynamics despite no architectural rank constraint, placing task representations in disparate subspaces of the shared core network. A simple high-level policy over the learned options, optimized while the whole network is frozen, sequences the low-rank adapters to produce novel out-of-distribution movements. We demonstrate the ability to generalize to novel motor sequences within the closed-loop control setting, improving on the generalization error of a task-input-conditioned multitask baseline by up to order of magnitude.

We propose MOSEL (Multi-Objective Stackelberg Efficient Learning), a framework for a posteriori multi-objective optimization (MOO) in deep neural networks that recovers a full front of Pareto stationary solutions at the computational cost of standard single-objective training. MOSEL reformulates the problem as a bilevel optimization problem that leverages network modularity to decouple representation learning from objective-preference alignment. Casting the bilevel problem as a Stackelberg game enables solving the original a posteriori MOO problem in a single forward–backward pass. As a result,MOSEL matches the time and memory efficiency of standard single-objective training while enabling scalable Pareto stationary front learning. Empirically, MOSEL uncovers diverse and optimal Pareto frontiers in strongly conflicting settings (e.g., fairness–accuracy). Remarkably, even in weakly conflicting regimes such as multi-task learning, it consistently converges to solutions closer to the utopia point, outperforming both standard single-objective training and specialized multi-task learning methods. These results highlight the broader potential of a posteriori MOO learning as a pathway to efficiently learn more diverse and robust representations, ultimately improving generalization.

Graph Neural Networks (GNNs) operate under the implicit assumption that the underlying data manifold is Riemannian, where the metric tensor is strictly positive-definite. While effective for homophilic graphs, this inductive bias creates a fundamental geometric mismatch for heterophilic graphs, where edges often signify dissimilarity or structural repulsion. In this work, we propose a paradigm shift from learning on fixed manifolds to learning the manifold itself. We postulate that heterophilic graphs are naturally embedded in pseudo Riemannian manifolds endowed with an indefinite metric, allowing for negative squared distances to model repulsive interactions. To formalize this, we introduce the Graph Signature Index, a spectral invariant that diagnoses the geometric nature of a graph. This index enables our proposed Pseudo-Riemannian Attention Network (PRAT) to dynamically learn an indefinite metric tensor, effectively capturing both attractive and repulsive interactions. By unifying these opposing forces within a single framework, PRAT demonstrates how a subtle shift in the metric signature yields substantial performance gains on heterophilic benchmarks. Our analysis reveals that PRAT’s learned geometry emergently aligns with the spectral signature of the graph, validating our geometric hypothesis.

Neural network verification is key to certifying robustness in safety-critical systems. Sound verifiers commit to abstract domains that propagate soundly through the network's operations, paying for that commitment with looseness at scale and applicability largely confined to piecewise-linear architectures. Probabilistic verifiers relax soundness, but they too commit to a fixed family of shapes for the network's reachable outputs. When the network's true output geometry differs from the chosen shape, the verifier over-approximates the reach-set and fails to certify networks that are in fact safe. We propose instead to learn the reach-set geometry from the network's behavior. A flow-matching model trained on input-output samples and calibrated by conformal prediction yields a probabilistic reach set whose shape adapts to the true shape of the reach-set rather than to an a priori choice. The resulting set has no closed-form description, so we reframe the specification check as a rare-event estimation problem on the learned distribution, restricted to the calibration's coverage region, and solve it with bounded adaptive multilevel splitting. The pipeline produces a probabilistic safety certificate combining the conformal coverage with a rare-event upper bound. Our pipeline performs comparably to sound verifiers on a subset of VNN-COMP 2025 benchmarks where sound verification is tractable and matches or exceeds fixed-shape probabilistic baselines on most benchmarks in the same suite. On a synthetic family of growing-depth networks, it scales sub-exponentially where sound verifiers time out or abstain.

Animals generate complex behaviors by flexibly recombining a finite set of motor primitives, but existing behavior segmentation methods oversimplify this process by imposing discrete syllables under restrictive generative assumptions. Here, we introduce Motif-based Continuous Dynamics (MCD), a framework that models behavior as a continuous, compositional process driven by reusable motor motifs. MCD leverages reinforcement learning (RL) to (1) discover interpretable motif representations via transition-based representation learning, and (2) model behavior as a time-varying mixture of these motifs through motif-based policies. This formulation avoids restrictive dynamics assumptions while capturing continuity, compositionality, and long-term dependencies in behavior. Across simulated gridworld tasks, maze navigation, and animal behavior datasets, MCD identifies reusable motifs and interprets the trajectories with them. These results provide a generative account of behavior as combinations of fundamental motifs, offering a flexible and interpretable framework for studying natural behavior.

Large language model (LLM)-based coding agents achieve strong results on controlled benchmarks yet routinely produce pull requests that real maintainers reject. The cause is not functional incorrectness but a lack of organicity: generated code ignores project conventions, duplicates internal APIs, and violates implicit architectural constraints. The latest repository snapshot is insufficient-it reveals the final codebase state, not the change patterns that shaped it. We make three contributions. (1) Paradigm: we propose next-commit prediction as a paradigm for learning agentic coding skill from a repository's own history-each historical commit is a self-supervised target whose oracle diff supplies dense supervision. (2) Evaluation: we establish Organicity Evaluation as a measurable objective for coding agents and contribute the Learning to Commit benchmark-5 curated open-source GitHub repositories under strict per-repository temporal splits-scoring patches along file localisation, internal API reuse, patch bloat, and code-style consistency. (3) Method: we instantiate the paradigm with the Learning to Commit framework, in which the agent performs supervised contrastive reflection: it blindly attempts each historical commit, contrasts its prediction against the oracle, and incrementally distils a reusable skill document that conditions subsequent generations. On the Learning to Commit benchmark and on SWE-bench Pro reframed under our temporal-split protocol, our framework consistently improves organicity on held-out future tasks and further lifts test-pass rates on SWE-bench Pro, narrowing the gap between benchmark success and real-world mergeability.

Self-supervised monocular depth estimation is appealing for its potential to scale with unlabeled data, yet existing methods often degrade when trained on diverse scenes. This study finds that, under self-supervision, models tend to encode dataset-specific structural biases rather than transferable scene geometry, leading to cross-domain optimization conflicts. To address this issue, this study proposes the MaSS framework, aiming to make self-supervised depth learning scalable to diverse data. MaSS introduces a dedicated structural representation pathway explicitly decoupled from contextual features, which is guided by a novel scene-structure discrimination objective to learn generalized, layout-aware structural representations for depth decoding. Trained and evaluated on four diverse datasets, both individually and jointly, MaSS substantially outperforms prior self-supervised methods and is the first to consistently benefit from mixed-domain training. Zero-shot evaluation on eight additional datasets further demonstrates its remarkable generalizability, particularly on those entirely out-of-distribution scenes. The source code will be publicly available upon publication.

Prompt injection is a critical vulnerability in LLM agents, yet the strongest methods still rely on human red-teamers and hand-crafted prompts. Adapting automated jailbreak optimizers does not close this gap: jailbreaks shape models toward generic compliance, while prompt injection requires emitting specific tool calls with correct parameters. The success signal is binary, and randomly sampled suffixes almost never trigger it—so standard optimizers have no gradient to follow. We present AutoInject, a black-box reinforcement learning (RL) framework that learns adversarial suffixes for prompt injection. A learned comparison-based reward scores each candidate against the best suffix seen so far, turning the binary signal into a dense reward suitable for RL optimization. The framework supports both online query-based attacks and offline-trained transferable suffixes that need no utility access at deployment, and incorporates a utility objective when task-completion feedback is available. On AgentDojo, AutoInject outperforms template attacks, GCG, TAP, and adaptive attack across production models, with statistically significant improvements under McNemar's test with $p<0.05$. Suffixes learned by AutoInject also break Meta-SecAlign-70B, a model fine-tuned specifically to resist prompt injection, where template attacks fail outright. The results establish an automated baseline for prompt injection and expose a gap between preference-based defenses and adaptive optimization-based attackers.

Boolean satisfiability (SAT) provides a foundation tool for cryptographic key recovery by encoding ciphers into CNF or ANF representations. However, heuristic solvers and their neural-enhanced variants often degrade significantly on cryptographic instances due to pronounced structural symmetry, a large number of intermediate variables, and complex algebraic dependencies. Existing neural approaches—ranging from end-to-end prediction to solver-integrated heuristics—face challenges in scalability, assignment accuracy, or computational efficiency. In this paper, we propose a unified neural-guided enumerative SAT framework for cryptographic key recovery. It performs a single forward neural guidance to identify a subset of $k$ critical variables, followed by enumeration over their assignments before invoking a heuristic SAT solver. This design effectively reduces the combinatorial search space while keeping neural overhead minimal. We further introduce a taxonomy of neural guidance across data and model capability regimes, including supervised selection with cryptographic priors, uncertainty-driven selection with assignment prediction, and robust fallback strategies using the model trained on general SAT datasets. Experiments on 10 SAT solvers and 2 cube-and-conquer strategies over SAT4CryptoBench and SAT Competition benchmarks demonstrate up to $5\times$ speedup and approximately 2× average improvement on cryptographic instances, while maintaining better performance on general SAT datasets. These results highlight the effectiveness and generalization of this framework, illustrating its potential to bridge machine learning and SAT solving in cryptanalysis tasks.

Synthetic data has become a promising way to scale model training beyond limited human-generated data but it may also induce strong model collapse (Dohmatob et al., 2024), where any fixed fraction of synthetic data prevents model performance from improving under data scaling, leaving a non-vanishing excess risk floor. In this paper, we studied how synthetic data affects the generalization of one-pass SGD in high-dimensional linear regression with model shift. We show that mixed training induces strong model collapse while two-stage training avoids this by using synthetic data only in the first stage, followed by real-data training in the second stage, showing that strong model collapse is not inevitable through a simple data curriculum. We further establish scaling-law upper bounds for both protocols under a random sketch model, showing that larger models amplify synthetic-induced degradation in mixed training and giving an explicit characterization of how high-quality synthetic training may reduce bias in two-stage training. Overall, our results highlight that synthetic data is neither inherently harmful nor beneficial; its effect depends critically on both its quality and the training protocol used to incorporate it.

Optical lens design is a complex, non-convex optimization challenge that relies heavily on human experience and intuition. Existing optimized-based automatic lens design methods struggle to navigate this vast parameter space without meticulous manual tuning. In this paper, we present LensDesigner, an autonomous agent framework that mirrors the problem-solving workflow of expert opticians. To overcome the initial cold start problem, we construct LensLib100K, an extensive optical lens library, and employ Optics-Aware Retrieval to supply physically valid structural seeds. Within an interactive physical simulation environment, the agent executes macroscopic orchestration while receiving immediate optical feedback. Furthermore, we introduce a continuous self-evolving mechanism guided by a curriculum agent. By iteratively solving design tasks with progressively increasing difficulty, the agent autonomously extracts, accumulates, and reuses design heuristics, effectively evolving its optical lens design expertise over time. At the evaluation level, we introduce LensArena, a standardized evaluation benchmark comprising 120 diverse optical design tasks, covering extreme configurations. Extensive experiments on this benchmark demonstrate that LensDesigner significantly outperforms publicly available baseline algorithms, achieving superior success rates and optimization efficiency. We hope this work sheds light on the emerging field of intelligent optics. The code will be publicly available.

Distilled diffusion models accelerate image generation by reducing the number of denoising steps, but often suffer from degraded image quality. To mitigate this trade-off, test-time optimization methods improve quality, yet their iterative nature incurs substantial computational overhead and leads to slow inference, limiting practical usability. Recent hypernetwork-based approaches amortize this process during training, but still require costly noise modulation in high-dimensional latent spaces. In this work, we propose LENS (Low-frequency Eigen Noise Shaping), an efficient noise modulation framework that operates in a low-dimensional subspace. Our approach is motivated by the observation that low-frequency components of the noise largely determine the global structure and visual fidelity of generated images. Based on this observation, we provide a theoretical justification for restricting modulation to the low-frequency subspace and derive a principled training objective. Building on this, LENS employs a lightweight, standalone network to selectively modulate these components, enabling efficient and targeted noise modulation. Extensive experiments demonstrate that LENS achieves competitive image quality while reducing FLOPs by 400–700$\times$, model parameters by 25–75$\times$, and inference-time overhead by 10–20$\times$ compared to prior methods.

Vision Language Models (VLMs) offer the exciting possibility of processing text as rendered images, bypassing the need for tokenizing the text into long token sequences. Since VLM image encoders map fixed-size images to a fixed number of visual tokens, varying rendering resolution provides a fine-grained compression knob. However, accuracy deteriorates quickly as compression increases: characters shrink below the vision encoder's effective resolution, making them indistinguishable. To address this, we propose LensVLM, an inference framework and post-training recipe that enables VLMs to scan compressed images, then selectively expand only the relevant images to their uncompressed form via learned tools. Building on Qwen3.5-9B-Base, LensVLM maintains accuracy comparable to the full-text upper bound at 4.3$\times$ effective compression and outperforms retrieval-based, text- and visual-compression baselines up to 10.1$\times$ effective compression across seven text QA benchmarks. LensVLM also generalizes to multimodal document and code understanding tasks, with the accuracy gain over baselines growing as compression increases. Our analysis validates this approach: training makes visual compression robust to rendering choices, and as compression grows the model increasingly relies on expanded content rather than unreliable visual reading. The analysis also yields practical tool-choice guidance: text expansion is preferable for rendered text, while high-resolution image expansion suits native documents whose layout cues carry task-relevant information.

Evidence-grounded medical question answering requires not only accurate answers, but also precise, verifiable, and clinically trustworthy evidence. Existing medical RAG systems typically retrieve top-(k) chunks according to item-wise relevance. Although this design improves recall, it does not explicitly optimize whether the selected evidence set is compact, sufficient, and decision-critical. In this paper, we study medical evidence-grounded QA from a different perspective: instead of retrieving evidence without optimization, the evidence-grounded medical QA system should identify a compact sufficient subset of high-density evidence snippets. We propose MEGA, a gain-aware framework for compact supporting medical evidence selection. MEGA first expands clinical queries into complementary subqueries and performs hybrid retrieval to construct a recall-oriented candidate evidence pool. It then introduces hidden-state Information Gain Scoring, which uses a frozen LLM to estimate whether each candidate snippet contributes new answer-relevant information beyond surface relevance. Finally, MEGA formulates evidence selection as a budgeted utility maximization problem and proposes an FPTAS selector to identify a compact evidence subset under an explicit token budget. We further construct two evidence-grounded medical QA benchmarks, CRC-EvidenceQA and Med-EvidenceQA, to evaluate both answer quality and evidence grounding. Across three LLM backbones and five medical RAG baselines, MEGA consistently achieves the best results, improving over the second-best baseline by (8.8\%) on average in answer quality and 38.8\% on average in evidence F1. Ablation studies and blinded clinical expert evaluation further validate its robustness and clinical relevance.

Vision-language-action models (VLA) promise human-steerable autonomous driving, but their training is bottlenecked by the scarcity of frames paired with natural-language instructions: while camera streams and expert trajectories are logged at scale, language annotations (e.g., "turn left at the intersection") remain scarce and expensive to acquire. To address this challenge, we introduce Latent Action Driving Annotations (LADA), a three-stage pipeline that transforms abundant unlabelled observation–trajectory pairs into a substrate for language-conditioned control. First, we train a latent action model with a vector-quantised bottleneck, producing a compact codebook of high-level vehicle intents. Second, a small language-annotated subset is used to train a vision-language translator to map observations and language instructions into this codebook. Third, we train a driving VLA on observation–latent-action pairs over the full unlabelled corpus. Using fewer than 5% of language annotations and without leveraging any auxiliary chain-of-thought reasoning or visual question answering streams, LADA achieves a Driving Score of 87.98 and a Success Rate of 70.46% on the closed-loop Bench2Drive benchmark, matching or surpassing fully supervised baselines.

Localizing AI-edited regions is essential for interpretable forensic analysis, but remains challenging due to subtle and spatially distributed artifacts that are misaligned with semantic or object boundaries. Existing approaches rely on pixel-level supervision from controlled editing pipelines, which is difficult to scale and can introduce misleading signals: artifacts frequently extend beyond annotated regions, while out-of-mask pixels are treated as authentic. This limits models' ability to capture transferable evidence and generalize across generators and datasets. To address these issues, we propose \textbf{ReGFLoW}, a \textbf{Re}construction-\textbf{G}uided \textbf{F}ake \textbf{Lo}calization framework under \textbf{W}eak supervision, which is the first weakly supervised approach for diffusion-edited fake region localization. ReGFLoW requires only real/fake labels at the image-level and uses diffusion reconstruction errors as dense spatial guidance to inject them into both feature and score spaces. Furthermore, by artifact-centric multiple instance learning, ReGFLoW utilizes localized diffusion evidence without relying on semantic-affinity or boundary-based pseudo-mask priors. Extensive experiments demonstrate that ReGFLoW achieves stronger out-of-domain generalization than fully supervised learning baselines.

Currently, there is a gap in the field of ultra-high-definition (UHD) video dehazing due to the lack of a benchmark for evaluation. Furthermore, existing video dehazing methods cannot run on consumer-grade GPUs when processing continuous UHD sequences of 3--5 frames at a time. In this paper, we address both issues with a new benchmark and an efficient method. Our key observation is that atmospheric dehazing reduces to a per-pixel affine transform governed by the low-frequency depth field, which can be compactly encoded in bilateral grids whose prediction cost is decoupled from the output resolution. Building on this, we propose LiBrA-Net, which factorizes the spatiotemporal affine field into a spatial--color and a temporal bilateral sub-grid predicted at a fixed low resolution, fuses their coefficients in the $\mathfrak{gl}(3)$ Lie algebra under group-theoretic regularization, maps the result to invertible $GL(3)$ transforms via a Cayley parameterization, and restores high-frequency detail through a lightweight input-guided branch. We further release UHV-4K, the first paired 4K video dehazing benchmark with depth, transmission, and optical-flow annotations on every frame. Across UHV-4K, REVIDE, and HazeWorld, LiBrA-Net sets a new state of the art among compared video dehazing methods while running native 4K at 25\,FPS on a single GPU with only 6.12\,M parameters. Code and data are available at \url{https://anonymous.4open.science/r/LiBrA-Net-42B8}.

With the rapid development of Video Large Language Models (VLLMs), online video understanding has emerged as a practical paradigm for real-world streaming applications. A key challenge is to construct a compact, query-agnostic visual memory from continuously growing video streams under strict token budgets. Existing token reduction methods typically rely on static and coarse-grained importance estimation, implicitly assuming that retained tokens contribute equally over time. This overlooks the dynamic nature of visual information, where the relevance of tokens evolves as the video progresses, leading to suboptimal prioritization and inefficient long-term memory usage. To address this limitation, we propose LifeStream, a training-free framework for lifecycle-aware visual token selection and hierarchical memory construction. LifeStream models the temporal evolution of visual tokens and reveals that tokens contribute unequally to downstream reasoning depending on their lifecycle states. Based on this insight, it organizes tokens into a hierarchical memory structure: informative non-stable tokens are preserved in an archive memory, while long-range history is progressively compressed into a global memory according to state-specific priorities. This design enables continuous retention of high-value information under limited token budgets, without requiring query-time retrieval or additional computation. Extensive experiments demonstrate that LifeStream discards over 80\% of visual tokens while improving performance. It improves LLaVA-OV-7B and Qwen3-VL-8B by 4.4\% and 3.5\% on StreamingBench, respectively, and achieves state-of-the-art results on one additional online benchmark and four long-video understanding benchmarks.

Flow-based vision-language-action (VLA) models provide expressive action generation for embodied control, but supervised fine-tuning (SFT) often confines them to narrow expert behaviors. Online reinforcement learning (RL) can improve beyond demonstrations through environment interaction, yet flow-based VLAs struggle with sparse long-horizon rewards and intractable action likelihoods. We propose \textbf{\emph{LiFT}}, a likelihood-free tree policy optimization framework for flow-based VLAs. LiFT first addresses sparse-reward credit assignment by creating sibling continuations from shared rollout histories and using their final outcomes to compute branch level relative advantages. LiFT further uses local action-flow instability as an expansion criterion, focusing rollout budget on ambiguous action-generation regions that are most likely to reveal meaningful outcome differences. Finally, LiFT applies branch-level advantages through a decision-level surrogate ratio on executed action chunks, matching policy updates to the granularity of tree-based credit assignment. This yields a likelihood-free policy update without attributing chunk-level credit to individual denoising steps. Evaluations on in-distribution and out-of-distribution benchmarks show that LiFT improves task success and robustness to distribution shifts.

The extension of context windows in Large Language Models is typically facilitated by scaling positional encodings followed by Continued Pre-Training (CPT). While effective, this paradigm is notoriously data-hungry and computationally expensive, requiring massive long-text corpora to recalibrate the model to the shifted positional distribution. We propose LinearARD, a self-distillation method that restores Rotary Position Embedding (RoPE)-scaled students through attention-structure consistency with a frozen native-RoPE teacher. Rather than next-token prediction or opaque hidden-state matching, LinearARD aligns row-wise distributions of dense $Q/Q$, $K/K$, and $V/V$ self-relation matrices from the final attention layer to directly supervise attention dynamics. To remove the quadratic memory bottleneck of $n \times n$ relation maps, we introduce a linear-memory kernel that stores only per-token log-sum-exp statistics and recomputes logits in the backward pass to obtain exact Kullback-Leibler divergence gradients. Across LLaMA2-7B, LLaMA3-8B, and Mistral-7B-v0.1 extended to 32K context, LinearARD recovers 93.1\%/94.2\%/94.3\% of native short-context performance and achieves strong long-context robustness on RULER using only \textbf{4.25M} tokens---amounting to just 1.6\% (a $\sim$60$\times$ reduction) of the 256M-token budget required by state-of-the-art baselines. Under this severely constrained budget, CPT and LongReD remain near zero on RULER, demonstrating that relation-level restoration provides a substantially better efficiency-quality tradeoff.

Interpreting the concepts encoded by individual neurons in deep neural networks is a crucial step towards understanding their complex decision-making processes and ensuring AI safety. Despite recent progress in neuron labeling, existing methods often limit the search space to predefined concept vocabularies or produce overly specific descriptions that fail to capture higher-order, global concepts. We introduce LINE, a novel, training-free iterative approach tailored for open-vocabulary concept labeling in vision models. Operating in a strictly black-box setting, LINE leverages a large language model and a text-to-image generator to iteratively propose and refine concepts in a closed loop, guided by activation history. We demonstrate that LINE achieves state-of-the-art performance across multiple model architectures, yielding AUC improvements of up to 0.11 on ImageNet and 0.05 on Places365, while discovering, on average, 27% of new concepts missed by massive predefined vocabularies. Beyond identifying the top concept, LINE provides a complete generation history, enabling polysemanticity evaluation and producing supporting visual explanations that rival gradient-dependent activation maximization methods. The LINE source code is available on GitHub.

Large language models (LLMs) routinely fail to output the correct option in multiple-choice question answering (MCQA) while encoding the answer internally. We expose this latent knowledge via the Query–Key (QK) score, defined for an attention head as the inner product between the last-token query and the key at the end-of-line token following option $i$, evaluated before rotary positional embedding is applied. Its argmax identifies a universal class of select-and-copy heads in middle layers that perform option selection through semantic query–key alignment, mechanistically distinct from induction and copy-suppression heads (Olsson et al., 2022): they are invariant to label symbols, and solve a synthetic task with zero surface overlap—properties no positional-copy account explains and that critically require stripping RoPE. Across $24$ models from $1.5\mathrm{B}$ to $72\mathrm{B}$ parameters (LLaMA-2/3/3.1/3.3, Qwen-2.5, Gemma, Phi-3.5, DeepSeek-R1-Distill), a single head's QK-score exceeds the model's own zero-shot accuracy by up to $+27.4$ pp on HellaSwag and $+49.8$ pp on HaluDialogue; causal zero-ablation collapses MCQA accuracy to near-random. To remove any dependence on labeled validation data, we introduce an unsupervised HeadScore that ranks heads from unlabeled inputs and recovers the supervised top-$k$ heads on every tested model. Against four positional-debiasing baselines (e.g., PriDe, Wiegreffe, Wang), QK-score is complementary by construction: debiasing re-weights output logits, whereas QK-score reads the model's selection from a middle-layer head before decoding. We release a one-line drop-in $\mathtt{HeadScore}$ script and per-model head indices, making every result one-command reproducible across all $24$ models and four benchmarks.

Reinforcement learning with verifiable rewards (RLVR) has become a standard approach for large language models (LLMs) post-training to incentivize reasoning capacity. Among existing recipes, group-based policy gradient is prevalent, which samples a group of responses per prompt and updates the policy via group-relative advantage signals. This work reveals that these optimization strategies share a common geometric structure: each implicitly defines a target distribution on the response simplex and projects toward it via first-order approximation. Building on this insight, we propose Listwise Policy Optimization (LPO) to explicitly conduct the target-projection, which demystifies the implicit target by restricting the proximal RL objective to the response simplex, and then projects the policy via exact divergence minimization. This framework provides (i) monotonic improvement on the listwise objective with bounded, zero-sum, and self-correcting projection gradients, and (ii) flexibility in divergence selection with distinct structural properties through the decoupled projection step. On diverse reasoning tasks and LLM backbones, LPO consistently improves training performance over typical policy gradient baselines under matched targets, while intrinsically preserving optimization stability and response diversity. The code is available at https://anonymous.4open.science/r/LPO.

Modern language models are trained on heterogeneous web-scale text corpora. Consequently, studying knowledge and skill acquisition is difficult, as prior exposure to related content is hard to characterize. To address this challenge, we introduce LittleCurriculum, a meticulously curated 88B-token pretraining corpus tailored to U.S. elementary school material, explicitly excluding concepts, facts, and vocabulary taught above Grade 5. Training a 2B-parameter LLM from scratch on LittleCurriculum yields LittleLearner, a model with sufficient language competence for open-ended evaluation, yet with clear knowledge and capability boundaries mapped to interpretable curriculum guidelines. We release LittleCurriculum and LittleLearner as a developmentally restricted sandbox to study how models acquire, represent, and use data under a well-defined training scope. We illustrate the sandbox's utility in a first suite of experiments on injecting new knowledge through post-training and in-context learning. These methods let LittleLearner better utilize existing knowledge, but do not raise out-of-scope capabilities. Our findings underscore the value of this controlled environment for future investigations.

We develop a game-theoretic framework for predicting and steering the behavior of populations of large language models (LLMs) through Nash equilibrium (NE) analysis. To avoid the intractability of equilibrium computation in open-ended text spaces, we model each agent’s action as a mixture over human subpopulations. Agents choose actively and strategically which groups to align with, yielding an interpretable and behaviorally substantive policy class. We derive closed-form NE characterizations, adopting standard concave-utility assumptions to enable analytical system-level predictions and give explicit, actionable guidance for shifting alignment targets toward socially desirable outcomes. The method functions as an active alignment layer on top of existing alignment pipelines such as RLHF. In a social-media setting, we show that a population of LLMs, especially reasoning-based models, may exhibit political exclusion—pathologies where some subpopulations are ignored by all LLM agents—which can be avoided by our method, illustrating the promise of applying the method to regulate multi-agent LLM dynamics across domains.

Tool-augmented LLM agents tend to call tools indiscriminately, even when the model can answer directly. Each unnecessary call wastes API fees and latency, yet no existing benchmark systematically studies when a tool call is actually needed. We propose When2Tool, a benchmark of 18 environments (15 single-hop, 3 multi-hop) spanning three categories of tool necessity — computational scale, knowledge boundaries, and execution reliability — each with controlled difficulty levels that create a clear decision boundary between tool-necessary and tool-unnecessary tasks. We evaluate two families of training-free baselines: Prompt-only (varying the prompt to discourage unnecessary calls) and Reason-then-Act (requiring the model to reason about tool necessity before acting). Both provide limited control: Prompt-only suppresses necessary calls alongside unnecessary ones, and Reason-then-Act still incurs a disproportionate accuracy cost on hard tasks. To understand why these baselines fail, we probe the models' hidden states and find that tool necessity is linearly decodable from the pre-generation representation with AUROC 0.89–0.96 across six models, substantially exceeding the model's own verbalized reasoning. This reveals that models already know when tools are needed, but fail to act on this knowledge during generation. Building on this finding, we propose Probe&Prefill, which uses a lightweight linear probe to read the hidden-state signal and prefills the model's response with a steering sentence. Across all models tested, Probe&Prefill reduces tool calls by 48% with only 1.7% accuracy loss, while the best baseline at comparable accuracy only reduces 6% of tool calls, or achieves a similar tool call reduction but incurs a 5× higher accuracy loss. On the real-world Search-o1 agentic benchmark, Probe&Prefill reduces API calls by 20–56% without accuracy degradation.

Identifying manifold structure reveals hidden signals from datasets, removing noise. In this paper, we study the problem of identifying local manifold structure from data. Given a query point $y_0$ from a dataset, our goal is to recover the geometry of the data manifold in a neighborhood of $y_0$ using a \emph{pretrained} optimal transport flow from the reference to the data distribution. First, we prove that the Brenier optimal transport map preserves manifold structure: the preimage of an $m$-dimensional data manifold is itself an $m$-dimensional manifold in the reference space. Second, motivated by this result, we propose a latent variable model that maps a linear model through the transport flow. We prove that the linear approximation error is significantly reduced by the optimal transport map, leading to a tight fit of the non-linear data manifold. Third, noting the intractability of the resulting likelihood, we deploy denoising Fisher score estimation --- a recent development from simulation-based inference that learns the Fisher score over parameter--observation pairs --- to perform likelihood-based inference effectively. Experiments on both synthetic and real-world datasets demonstrate the effectiveness of the proposed method.

Orthogonal parameter-efficient fine-tuning adapts pretrained weights through structure-preserving multiplicative transformations, but existing methods often conflate two distinct design choices: the subspace in which adaptation occurs and the transformation applied within that subspace. This paper introduces LOFT, a low-rank orthogonal fine-tuning framework that explicitly separates these two components. By viewing orthogonal adaptation as a multiplicative subspace rotation, LOFT provides a unified formulation that recovers representative orthogonal PEFT methods, including coordinate-, butterfly-, Householder-, and principal-subspace-based variants. More importantly, this perspective exposes support selection as a central design axis rather than a byproduct of a particular parameterization. We develop a first-order analysis showing that useful adaptation supports should be informed by the downstream training signal, motivating practical gradient-informed support selection strategies. Across language understanding, visual transfer, mathematical reasoning, and multilingual out-of-distribution adaptation, LOFT recovers principal-subspace orthogonal adaptation while gradient-informed supports improve the efficiency–performance trade-off under matched parameter, memory, and transform budgets. These results suggest that principled support selection is an important direction for improving orthogonal PEFT.

We study the $K$-armed logistic bandit problem, where at each round, the agent observes $K$ feature vectors associated with $K$ actions. Existing approaches that achieve a rate-optimal $\widetilde{\cal O}(\sqrt{dT})$ regret bound rely heavily on context diversity assumptions, such as strict positivity of the minimum eigenvalue of a context covariance matrix. These assumptions, however, impose strong restrictions on the context process, as they rule out the situation where the context vectors are concentrated in a low-dimensional subspace. In this paper, we propose SupSplitLog, which, to the best of our knowledge, is the first algorithm for logistic bandits that achieves $\widetilde{\cal O}(\sqrt{dT})$ regret without any context diversity assumption. The key idea is to split the collected samples into two disjoint subsets when constructing estimators; one is used to compute an initial-point estimator, while the other is used to apply a Newton-type one-step correction procedure. The splitting rule is carefully designed to balance the accuracy requirements of the initial-point estimator and the one-step correction procedure. Moreover, SupSplitLog strictly improves on the existing algorithms in terms of the dependence on dimension $d$ in the regret upper bound. Furthermore, SupSplitLog can be adapted simply to deduce a regret bound that grows with a data-dependent complexity measure, avoiding a direct dependence on $d$, which is favorable when the context vectors are concentrated in a low-dimensional subspace. We also provide experimental results that demonstrate numerically the superiority of our algorithm, validating the theoretical results.

Real-world human activities unfold as sequences of temporally dependent human-object interactions (HOIs), yet existing methods address either atomic HOIs in isolation or text-driven motion composition without object or scene context. To bridge this gap, we introduce the task of long-term action composition of HOIs, where a human navigates between and interacts with multiple objects within a 3D scene. The primary challenges lie in producing smooth transitions, scene-aware locomotion, and plausible hand-object interaction. To this end, we propose LT-HOI, a diffusion-based model that jointly conditions on past and future contexts—the past providing kinematic continuity across transitions, the future enabling anticipatory navigation toward upcoming interacting objects. We further integrate Diffusion Noise Optimization for collision avoidance and introduce a hand-object displacement loss to improve contact quality. On the ParaHome dataset, we establish the first benchmark for this task and demonstrate that LT-HOI effectively composes long, temporally coherent HOIs.

Instructional video editing has matured rapidly on short, single-shot clips, yet long videos, the dominant form of video in practice, remain largely unaddressed. Long videos exhibit failure modes that short-video editing does not stress-test: edit targets recur across temporally distant shots under varying viewpoint, scene, and scale, demanding cross-shot semantic consistency, while non-target intervals must be left unchanged, demanding accurate temporal grounding of the instruction. We present a hierarchical framework that addresses these two axes with dedicated components: a planning agent that grounds the instruction and partitions the timeline into shot-aligned segments, a long-term anchor bank that maintains a sparse set of high-quality cross-shot anchors through critic-driven iterative refinement on top of a strong image editing prior, and an anchor-conditioned video editor obtained by lightly adapting a pretrained short-video editor through LoRA, with no architectural changes. The editor is trained exclusively on short-video pairs and lifted to long videos at inference through its conditioning interface, requiring no long-video supervision. To evaluate this regime, we introduce LVEdit-Bench, the first benchmark designed for long-video instructional editing in the wild, together with a three-tier evaluation protocol that isolates per-shot fidelity, cross-shot identity preservation, and robustness to target-absent content under a structured VLM-judge protocol. Across all three settings, our method matches strong short-video editors on per-segment quality and substantially improves cross-shot consistency and grounding, with the margin widening as temporal complexity grows.

Despite the remarkable progress of large Vision-Language Models (VLMs) in Radiology Report Generation (RRG), existing black-box autoregressive generation paradigms lack intermediate clinical logical reasoning. This deficiency renders such models prone to generating linguistically fluent reports with erroneous diagnostic findings, which manifest as plausible hallucinations and thereby undermine clinical reliability. To this end, we propose a novel RRG paradigm termed "Look-Before-You-Leap", aimed at enhancing the clinical logical reasoning capabilities of the model to improve diagnostic accuracy. Specifically, we first construct a self-evolving Radiological Skill (RadSkill) repository to guide the teacher model in distilling high-confidence explicit clinical reasoning, and further internalize this reasoning capability into the implicit diagnostic intuition of the RRG baseline. Building upon this foundation, a novel preference optimization method termed Psychometric Preference Optimization (PSY-PO) is proposed. Grounded in Item Response Theory (IRT) from psychometrics, this method precisely samples clinically misleading dispreferred reports for preference optimization, thereby effectively mitigating plausible hallucinations. Extensive experimental results demonstrate that our proposed method significantly enhances both the generation quality and trustworthiness of radiology reports.

Masked diffusion models (MDMs) have emerged as a promising alternative to autoregressive models for language modeling, yet the effective design of transformer architectures for MDMs remains underexplored. In this paper, we show that selectively looping the early-middle transformer layers significantly improves both training efficiency and model performance in MDMs. We call this approach \textbf{LoopMDM} (Looped Masked Diffusion Model), which brings two key benefits: looping layers at training-time yields a depth-scaling effect without adding parameters, while varying the number of loops at inference-time enables flexible compute scaling. Despite the simplicity, the results are striking: across multiple pre-training corpora, LoopMDM matches the performance of same-size MDMs with up to $3.3\times$ fewer training FLOPs, while its final performance outperforms them on various reasoning benchmarks, including up to $+8.5$ points on GSM8K. It even surpasses deeper non-looped MDMs trained with comparable per-step compute, indicating that selective looping is more effective than naive depth scaling. Furthermore, LoopMDM can scale inference-time compute by increasing the number of loops. Adaptively adjusting the number of loops throughout the sampling process further yields additional gains in compute efficiency while maintaining performance. Lastly, with attention analysis, we provide evidence that looping is effective in MDMs by promoting interactions among masked positions. Our code and weights will be publicly released.

Large Language Models face a challenging multi-objective coupled optimization problem in constrained long-text generation, requiring the simultaneous balancing of global structure, local coherence, and constraint satisfaction. However, existing methods often rely on static pipelines or loosely coupled designs, making it difficult to achieve cross-stage collaborative optimization. To address this, we propose LoopWeaver, which models the generation process as a hierarchical, feedback-driven closed-loop optimization problem; by introducing iterative feedback between the planning and generation stages, it achieves the joint optimization of both global and local aspects. Specifically, LoopWeaver integrates constraint-aware hierarchical planning, feasibility filtering, and reward-based preference optimization, thereby allowing feedback signals to permeate both the planning and generation layers. Unlike methods that utilize feedback solely during post-processing or within a single module, LoopWeaver introduces a unified paradigm for feedback-coupled optimization, demonstrating significant improvements in text quality and constraint satisfaction capabilities across various models.

Low-Rank Adaptation (LoRA) has become a widely adopted technique in text-to-image diffusion models, enabling the personalisation of visual concepts such as characters, styles, and objects. However, existing approaches struggle to effectively compose multiple LoRA adapters, particularly in open-ended settings where the number and nature of required skills are not known in advance. In this work, we present LoRAtorio, a novel train-free framework for multi-LoRA composition that gates each adapter using only signals already inside the denoiser, specifically, patch-level cosine (dis)similarity between each LoRA's predicted noise and the base model's. We refer to this as ``intrinsic'' guidance because no external parameters are introduced. Our method is motivated by two key observations: (1) LoRA adapters trained on narrow domains produce unconditioned denoised outputs that diverge from the base model, and (2) when conditioned out of distribution, LoRA outputs show behaviour closer to the base model than when conditioned in distribution. These patch-level similarities are used to construct a spatially-aware weight matrix, which guides a weighted aggregation of LoRA outputs. To address domain drift, we further propose a modification to classifier-free guidance that incorporates the base model's unconditional score into the composition. We extend this formulation to a dynamic module selection setting, enabling inference-time selection of relevant LoRA adapters from a large pool. On the ComposLoRA benchmark, LoRAtorio achieves state-of-the-art performance, with a CLIPScore that does not deteriorate as the number of composed adapters grows, surpassing the strongest prior method by up to 1.3% in CLIPScore, and reaching a 76.92% GPT-4V win rate against the closest baseline. We further show that the framework is backbone-agnostic: the intrinsic-similarity principle transfers to rectified-flow architectures (FLUX.1-dev) without retraining. Code will be made available.

Training data attribution (TDA) identifies which training examples most influenced a model's prediction. Influence function methods are a theoretically grounded family of TDA methods and exploit gradients. To overcome the scalability challenge arising from gradient computation, the most popular strategy is random projection (e.g., TRAK, LoGRA). However, this still faces two bottlenecks when scaling to large training sets and high-quality attribution: (i) storing and loading projected per-example gradients for all $N$ training examples, where query latency is dominated by I/O; and (ii) forming the $D \times D$ inverse Hessian approximation, which costs $O(D^2)$ memory. Both bottlenecks scale with the projection dimension $D$, yet increasing $D$ is necessary for attribution quality---creating a quality--scalability tradeoff. We introduce \textbf{LoRIF} (Low-Rank Influence Functions), which exploits low-rank structures of gradient to address both bottlenecks. First, we store rank-$c$ factors of projected per-example gradients rather than full matrices, reducing storage and query-time I/O from $O(D)$ to $O(c\sqrt{D})$ per layer per sample. Second, we use truncated SVD with the Woodbury identity to approximate the inverse Hessian term in an $r$-dimensional subspace, reducing memory from $O(D^2)$ to $O(Dr)$. On models from 0.1B to 70B parameters trained on datasets with millions of examples, LoRIF achieves up to 20$\times$ storage reduction and query-time speedup compared to LoGRA, while matching or exceeding its attribution quality. LoRIF makes gradient-based TDA practical at frontier scale.

Our aim is to develop a unified model for sign language understanding that performs both sign language translation (SLT) and sign–subtitle alignment (SSA). Together, these two tasks enable the conversion of continuous signing videos into spoken language text and also the temporal alignment of signing with subtitles -- both beneficial for practical communication, large-scale corpus construction, and educational applications. To achieve this, our approach is built upon three design choices: (i) a lightweight visual backbone that captures manual and non-manual cues from human keypoints and lip-region images while preserving signer privacy and enabling end-to-end training; (ii) a Sliding Perceiver mapping network that aggregates consecutive visual features into word-level embeddings to bridge the vision–text gap; and (iii) scaling training to large datasets -- BOBSL covering British Sign Language (BSL) and YouTube-SL-25 covering American Sign Language (ASL) -- to promote generalisation across languages and signers. With this multi-task, multilingual pretraining and strong model design, we achieve state-of-the-art results on BOBSL, How2Sign, Phoenix14T, and FLEURS-ASL SLT benchmarks, and also for SSA on BOBSL and WMT-SLT-SRF. Beyond these standard, signer-interpreted datasets, we show successful SLT on the natural signing data from the BSL-Corpus.

Retrieval-augmented generation (RAG) strategies have empowered large language models (LLMs) through integration with external knowledge sources, leading to more accurate, up-to-date, and contextually relevant outputs. Recently, graph-based RAG methods have gained attention for leveraging relational structures that support multi-hop reasoning and retrieval. However, existing approaches remain limited in their ability to explicitly model query-aware relevance signals during retrieval. In this paper, we propose LP-RAG, a graph-based framework for document RAG that casts retrieval as an inductive link prediction problem. In particular, LP-RAG first constructs a graph encoding semantic relationships among chunks (i.e., candidates for retrieval), which is then augmented with chunk-conditioned synthetic queries that emulate potential user questions. This enables self-supervised learning of query-chunk relevance without relying on real query annotations. Retrieval is performed by predicting links between unseen queries and chunk nodes. Notably, LP-RAG is model-agnostic and can leverage a broad class of link prediction methods. To demonstrate its effectiveness, we evaluate LP-RAG across diverse settings and benchmarks. Our empirical results show that LP-RAG consistently outperforms existing methods in both retrieval quality and downstream generation performance, while maintaining competitive efficiency relative to learnable graph-based RAG approaches.

We study online scheduling in a shared-resource system where tasks of different types arrive sequentially and compete for limited resources. Each task is modeled as an $H$-stage episodic Markov decision process, which captures the sequential resource-allocation decisions made within the task. At the system level, the learner must coordinate resource sharing across tasks to maximize the long-term ratio of cumulative reward to cumulative cost over $K$ tasks, without knowledge of the task distribution. The key challenges are coupled intra-task Markovian dynamics and inter-task ratio maximization, further complicated by unknown models and large or continuous state spaces. To address these challenges, we propose Contextual Optimistic Lyapunov Decomposition (COLD), a Lyapunov-driven optimistic learning algorithm. COLD converts the long-term ratio objective into a queue-adjusted per-task surrogate, thereby decoupling reward-cost optimization from statistical value estimation. It uses double-optimistic least-squares value iteration to estimate reward and cost values under linear function approximation, and a softmax policy to smooth the greedy update with only a controlled approximation error. We prove a regret bound of $\tilde{\mathcal{O}}(\sqrt{Md^3H^3T})$, where $M$ is the number of task types, $d$ is the feature dimension, $H$ is the episode horizon, and $T=KH$ is the number of interaction steps. The $\sqrt{T}$ dependence is order-optimal up to logarithmic factors, and the bound scales with the feature dimension rather than the size of the state space. Experiments on synthetic environments and an empirical KuaiRand-based MDP demonstrate the effectiveness of the proposed method.

Neural surrogate models for physical simulations are trained on discretized samples of continuous domains, where the induced empirical measure leads to uneven supervision, biasing optimization and causing spatial inconsistencies in physical fidelity. To mitigate this measure-induced bias, we propose M$^3$ (Multi-scale Morton Measure), a scalable framework that balances training measures by partitioning space according to physical variation and allocating supervision across multiple scales. Applied to three industrial-scale datasets with diverse discretizations, M$^3$ consistently improves predictions in the continuous physical domain, achieving up to 4.7$\times$ lower error in large-scale volumetric cases. These gains persist under aggressive subsampling (160M $\rightarrow$ 16M $\rightarrow$ 1.6M points), where M$^3$-trained models outperform those trained on higher-resolution data, reducing physics-weighted relative $L_2$ error by 3--4$\times$ and the corresponding MSE by up to 13$\times$. These results highlight data distribution as a key factor in operator learning and position M$^3$ as a scalable, data-efficient approach for physically consistent modeling.

Auction-based Federated Learning (AFL) provides a principled framework for incentivizing data owners and allocating decentralized training resources to data consumers. However, practical AFL systems face a coupled valuation and allocation challenge: data quality is uncertain before training, post-training feedback can be strategically distorted, and heterogeneous deadlines render late model updates ineffective for learning. We propose MainFL, an incentive-compatible intertemporal data valuation mechanism for robust auction-based federated learning. MainFL integrates three components: (i) a bounded mutual-information peer prediction rule that elicits truthful pre-training quality reports without requiring ground-truth labels; (ii) an intertemporal reliability calibration scheme that updates future valuations using post-session validation signals; and (iii) a deadline-induced layered clinching mechanism that allocates quality-weighted data while enforcing budget and feasibility constraints.We theoretically establish Bayes–Nash truthfulness for data owners, allocation feasibility and budget safety, and truthful residual-demand equilibrium for data consumers under standard clinching-auction regularity conditions. Empirical results on widely used federated learning benchmarks show that MainFL improves social welfare by up to 12.1% and model accuracy by up to 7.4% compared to state-of-the-art AFL baselines, while substantially reducing straggler-induced task timeouts.