Skip to yearly menu bar Skip to main content


Poster

Quantile Benchmarking Heterogeneous Web Corpora in Open LLM Pretraining

Kairong Luo ⋅ Zhenbo Sun ⋅ Xinyu Shi ⋅ Shengqi Chen ⋅ Bowen Yu ⋅ Yunyi Chen ⋅ Chenyi Dang ⋅ Tao H Tao ⋅ Hui Wang ⋅ Fangming Liu ⋅ Kaifeng Lyu ⋅ Wenguang Chen
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P173
Fully open pretraining now has increasingly reproducible recipes and accessible web-scale corpora. What remains missing is a principled way to merge scored open-web data. Corpora such as DCLM Baseline and FineWeb-Edu are built with different filters and scorers, so their raw scores are not directly comparable. Because web data is large, heterogeneous, and token-budget dominant, choosing an additional retained fraction is a consequential compute decision rather than a minor sampling detail. We formulate this setting as heterogeneous top-p mixing: choosing top-p retained fractions for each corpus along its score distribution under a fixed total token budget. We propose quantile benchmarking, a proxy-training framework that probes fixed score quantiles from each source and compares them through a shared downstream-utility metric. The resulting measurements reveal benchmark-dependent source orderings, large within-source utility variance, and a strong knowledge bias in current scorers. In particular, higher-score regions often degrade broader commonsense benchmarks such as PIQA and HellaSwag. We then fit benchmark-conditioned utility curves and derive top-p mixing cutoffs through a global utility threshold. 40B-token ablation experiments show that quantile-benchmarking-guided thresholds improve over both uniform mixing and heuristic top-p filtering. Finally, we apply the resulting recipe to train a 2B model over 2.2T tokens with a fully open recipe, reaching the Pareto frontier of parameter efficiency among compared fully open models of similar scale.
Show more
View full details
Poster

Scaling Genomic Language Modeling with Unified Corpus and Evaluation in Bacteria

Maciej Wiatrak ⋅ Samuel T Horsfield ⋅ Aaron Weimann ⋅ Ramon Viñas Torné ⋅ Maria Ntemourtsidou ⋅ David Abelson ⋅ Adam Dinan ⋅ John Lees ⋅ Maria Brbic ⋅ Andres Floto
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P165
Progress in genomic language modeling depends on pretraining resources that are both diverse and comprehensive, as well as benchmarks that evaluate models across biological scales from genes to whole genomes. This need is especially acute in bacteria, which are extremely abundant, diverse, and central to human health, ecology, and biotechnology. Yet resources for large-scale pretraining and evaluation in bacterial genomics remain limited, with existing frameworks focusing largely on eukaryotic settings. Here, we introduce BacCorpus, a genome-level resource for training genomic LMs comprising 20B proteins, including 4.6B unique sequences, from 7M genomes spanning more than 150,000 species, and BacBench, a comprehensive benchmark spanning 10 datasets and 5 tasks across gene-, system-, and genome-level tasks. We evaluate a diverse suite of genomic LMs on BacBench and additionally develop BacLM, a 350M-parameter model trained jointly on protein sequences and intergenic DNA to further investigate how bacterial genomes are best represented. Across tasks, we find that performance depends strongly on biological scale: methods leveraging protein sequence are competitive on gene-level prediction, whereas models that incorporate wider genomic context excel at problems which involve complex multi-gene interactions. Together, BacCorpus and BacBench provide a practical foundation for training and evaluating robust, genome-aware genomic language models in bacteria at an unprecedented scale.
Show more
View full details
Poster

Token-to-Token Alignment of Text Embeddings for Semantic Blending

Saar Huberman ⋅ Ron Mokady ⋅ Or Patashnik ⋅ Daniel Cohen-or
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P259
In text-conditioned generative models, images are represented and controlled through text prompts. In practice, images are generated from sequences of tokens derived from these prompts. However, the space of token sequences lacks a consistent structure: semantically similar images may correspond to sequences that differ in wording, ordering, and placement of concepts, while similar token sequences may encode very different semantics. This lack of structure makes it difficult to perform smooth transitions in this space, hindering applications such as image blending and continuous control of edits. We argue that this limitation stems not from the absence of semantic structure, but from misalignment between representations. To address this misalignment, we introduce Token-to-Token alignment, a framework that establishes explicit semantic correspondence between tokens across prompts. Our approach transforms prompts into a structured representation in which semantically corresponding concepts occupy compatible positions, and then aligns their token embeddings based on semantic similarity. Concretely, the method consists of two stages: a structural alignment that rephrases prompts into a shared structured form, followed by an embedding-level alignment that matches token representations across prompts. With this alignment in place, simple linear interpolation becomes a meaningful operation, producing smooth and coherent semantic transitions and enabling applications such as blending and continuous editing. Our results show that text embedding spaces in such models implicitly encode a continuous semantic structure that becomes accessible once representations are properly aligned, suggesting that semantic control can be achieved by organizing existing representations rather than modifying the generative model.
Show more
View full details
Poster

Thinking with Visual Primitives

Ruijie Lu ⋅ Yiyang Ma ⋅ Xiaokang Chen ⋅ Lingxiao Luo ⋅ Zhiyu Wu ⋅ Zizheng Pan ⋅ Ting Chen ⋅ Gang Zeng
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P278
Despite the remarkable progress in Multimodal Large Language Models (MLLMs), the prevailing Chain-of-Thought (CoT) paradigms remain predominantly confined to the linguistic space. While recent advancements have focused on bridging the Perception Gap through high-resolution cropping (e.g., Thinking with Images), they overlook a more fundamental bottleneck: the Reference Gap. The inherent ambiguity of natural language often fails to provide precise, unambiguous pointers to complex spatial layouts, leading to logical collapse in tasks requiring rigorous grounding. In this work, we introduce Thinking with Visual Primitives, a novel reasoning framework that elevates spatial markers (e.g., points and bounding boxes) to "minimal units of thought". By interleaving these visual primitives directly into the thinking process, our model can "point" while it "reasons", effectively grounding its cognitive trajectory in the physical coordinates of the image. Notably, our framework is built on a highly optimized architecture with extreme visual token efficiency. Despite its compact model scale and significantly lower image-token budget, our model achieves frontier-competitive performance on a focused suite of challenging visual QA tasks, matching or exceeding models such as GPT-5.4, Claude-Sonnet-4.6, and Gemini-3-Flash. This demonstrates a path toward more efficient and scalable System-2-like multimodal intelligence.
Show more
View full details
Poster

Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs

Naman Deep Singh ⋅ Maximilian Mueller ⋅ Amit Peleg ⋅ Francesco Croce ⋅ Matthias Hein
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P208
Unlearning in large language models (LLMs) requires precisely removing specific information from a pre-trained model, e.g. to delete private data or harmful knowledge acquired during pre-training. However, existing unlearning methods often fall short under thorough evaluation. This paper makes three contributions. First, we introduce JensUn, a novel unlearning approach which leverages the Jensen-Shannon Divergence (JSD) as the training objective for both forget and retain sets. We provide theoretical arguments why JSD outperforms KL-Divergence, and show in extensive experiments that JensUn achieves the best forget-utility trade-off and demonstrates strong resilience to benign relearning, outperforming existing methods. Second, for a precise unlearning evaluation, we introduce LPF, a curated dataset of less prominent facts that provides a realistic unlearning scenario. Third, to comprehensively test unlearning methods, we (i) employ an LLM as semantic judge instead of the standard ROUGE score, and (ii) evaluate worst-case unlearning over various paraphrases and input formats. Our improved evaluation reveals that many unlearning methods are less effective than previously thought.
Show more
View full details
Poster

Domination-Avoiding Learning Agents Cannot Collude

Noam Nisan ⋅ Emmanuel Zerah
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P107
An influential paper of Calvano et al. empirically demonstrated that Q-learning agents spontaneously collude when placed as sellers that compete on prices in a natural market model. More recent results of Fish et al. empirically demonstrated that similar collusion happens with commercial LLMs. We formally prove that such collusion can also happen with external-regret-minimizing agents. We identify a very general class of agents, which we term Domination-Avoiding agents, that provably do not collude in such markets. This class contains all Mean-Based agents and all internal-regret-minimizing agents, as well as others such as Multiplicative-Weight agents with variable learning rate and contextual variants thereof. More generally we show that, in any game, this class of agents is guaranteed to jointly learn to almost never play strategies that are eliminated by repeated elimination of purely dominated strategies.
Show more
View full details
Poster

Diff-CA: Separating Common and Salient Factors with Diffusion Models

Michaël Soumm ⋅ Alexandre Fournier Montgieux ⋅ Yunlong HE ⋅ Pietro Gori ⋅ Alasdair Newson
Dec 9, 10:30 PM - 12:30 AM Paris Poster Hall #P233
Contrastive Analysis aims to separate factors which are common between two data distributions from those which are salient to only one of them. Existing contrastive methods are based on generative models (e.g., VAEs or GANs) that often suffer from limited reconstruction and image quality, which hampers effective latent factor separation and limits their applicability to high-fidelity image generation and editing. We propose a novel conditioning framework for diffusion models that enables contrastive decomposition without compromising generation quality. We first train a prompt-free, image-conditioned diffusion model, and then learn to decompose the conditioning into a common and a salient factor, using weak supervision. We prove that the additive contrastive factorization, commonly assumed in prior work, is identifiable under mild conditions. This factorization enables targeted editing operations by swapping or interpolating only the salient factor while preserving the common factor.
Show more
View full details
Poster

SLAyiNG: A Diverse and Community-validated Dataset of Queer Slang

Leonor Veloso ⋅ Lea Hirlimann ⋅ Lucija M Zidar ⋅ Philipp Wicke ⋅ Valentin Hofmann ⋅ Hinrich Schuetze
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P201
Queer vernacular is rarely studied in NLP, despite advancements in resources and evaluation for other sociolects and informal language. Because of this, NLP systems often process queer language incorrectly, e.g., they misclassify it as hate speech or generate negative responses. To address this problem, we propose \textsc{Slaying}, the first real-world dataset of English queer slang. \textsc{Slaying} is community-validated, and includes over 400 queer slang terms that pertain to more than 20 queer subcommunities. We argue that queer language data resources have great potential in NLP -- e.g., as components of large pretraining corpora and as the basis for benchmarks -- and can improve queer users' experience of NLP systems. We leverage \textsc{Slaying} for two novel findings in support of this argument: (i) For a number of language models, we show that they are unbiased towards the queer community, but at the same time unable to process its language, i.e., absence of representation bias does not entail the absence of linguistic bias. (ii) Model performance on queer slang varies across queer subcommunities; it is generally worse for slang pertaining to African-American and Latine communities. These findings are relevant for both the queer NLP and the broader ML communities. \textsc{Slaying} is available to the public, and open to future revisions and extensions. \textbf{Warning:} This paper contains profane and potentially offensive language.
Show more
View full details
Poster

Doomed to Re-Annotate, Forever: The ImageNet Story

Illia Volkov ⋅ Nikita Kisel ⋅ Tetiana Mishkina ⋅ Klara Janouskova ⋅ Jiri Matas
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P318
Top-1 accuracy on ImageNet-1k remains the most universally reported metric in visual recognition, despite its well-documented label quality issues. In the paper, we describe the effort to obtain accurate and complete ImageNet-1k validation set annotations. The result -- ReImageNet -- is a from-scratch reannotation, which includes multilabel correction, object localisation, revised class definitions, and semantic attributes---text-recognition, rendition, reflection, crowd, and dominant. The reannotation reveals that $\approx13$\% of the original ImageNet-1k labels are incorrect, $32.7$\% of images are multilabel and $4.7$\% contain no object from an ImageNet-1k class. With the new labels, top-1 accuracy increases by $0.7$--$1.4$\% for supervised models and by $6$--$8$\% for MLLMs. We show that annotation errors in ImageNet-1k propagate into its derivative test sets, indicating that the problem is structural rather than specific to any single benchmark. We observed that LLMs outperform unaided crowd annotators, and that human and LLM collaboration with appropriate tooling represents the current quality ceiling for annotation at this scale. All [annotations](https://huggingface.co/datasets/c1rcuslegend/ReImageNet), [class definitions](https://gonikisgo.github.io/imagenet-annotation-preview/), [guidelines](https://gonikisgo.github.io/imagenet-annotation-preview/), and [analysis code](https://github.com/klarajanouskova/ImageNet) have been publicly released.
Show more
View full details
Poster

Complexity Aware Continuous Level of Details for Gaussian Splatting

Radu Beche ⋅ Raoul de Charette
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P260
3DGS enables real-time novel view synthesis, but practical deployment under varying compute budgets requires a single Gaussian set that remains effective when truncated to a prefix of primitives. This raises how capacity should be allocated as the budget decreases.We observe two findings. First, image complexity correlates with reconstruction error in both image and 3D Gaussian space. Second, this structure degrades under strong budget reduction, indicating suboptimal allocation.These observations motivate using image complexity as an optimization signal. We propose a continuous level-of-detail method based on 3DGS-MCMC that uses it for (i) relocation toward difficult regions during training and (ii) importance-based retention of simple and detailed regions at inference. A variance constraint on residuals between full- and reduced-budget renderings further enforces uniform error under compression. Our method yields smoother LOD degradation than prior work, with largest gains in low-budget regimes. Code will be released.
Show more
View full details
Poster

Mildly Overparameterized ReLU Networks on Orthogonal Data: Incremental Learning and Implicit Bias

James Town ⋅ Etienne Boursier ⋅ Ben Lewis ⋅ Matthias Englert ⋅ Ranko Lazic
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P248
The successful training of neural networks hinges on the use of first order optimization methods, yet the theoretical characterization of these methods remains incomplete. This is especially true in settings with mild overparameterization. In this work, we study the gradient flow dynamics of two-layer ReLU networks from small initialization with orthogonal training data. We prove the limiting flow converges to a saddle-to-saddle jump process as the initialization scale tends to zero, revealing an incremental learning phenomenon in which a new neuron activates at each saddle. This analysis recovers the known result of Dana et al. (2025) that the network interpolates the training data with high probability as soon as $m \gtrsim \log(n)$, where $m$ is the network width and $n$ is the number of training samples. This incremental process characterization also allows us to derive a novel implicit bias result: the learned interpolator has an $\ell_2$-norm scaling as $\sqrt{n}$, which is within a constant factor of the minimal-norm interpolator. More broadly, our work provides the first rigorous proof of an incremental learning process for ReLU networks, whilst suggesting mildly overparameterized networks can converge to interpolating solutions whose complexity is of the same order as that of the optimal interpolator.
Show more
View full details
Poster

Towards Convergence of PPO: An Approximate Descent Approach

Leif Döring ⋅ Daniel A Schmidt ⋅ Moritz Melcher ⋅ Sebastian Kassing ⋅ Benedikt Wille ⋅ Tilman Aach ⋅ Simon Weissmann
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P134
Proximal Policy Optimization (PPO) and it's variants are the most widely used policy-gradient methods in reinforcement learning, yet the role of its actor update mechanism is still poorly understood theoretically. In particular, standard PPO combines clipped surrogate gradients, multiple epochs of minibatch updates, and reuse of rollout data, but existing convergence analyses do not fully capture this update structure. In this work, we theoretically study PPO policy updates with symmetric clipping from a policy-gradient perspective and interpret its actor updates as a cyclic biased-gradient method with sample reuse and random reshuffling. Our first contribution is a clean formalization of PPO clipped actor updates through surrogate gradients that approximate the true policy gradient. Using the performance difference lemma, we prove a linear bias bound which quantifies how the surrogate gradient drifts as the policy moves away from the sampling policy. Our second contribution is a convergence analysis of cyclic surrogate-gradient ascent, showing that additional PPO-style biased updates can improve progress under conservative learning rates without requiring additional samples. Finally, we analyze the stochastic minibatch version with reshuffling and obtain convergence-to-stationarity guarantees under standard smoothness assumptions and a bounded critic-bias condition. Overall, our results provide a theoretical interpretation of PPO’s multi-epoch actor updates: extra clipped surrogate steps introduce bias, but can still improve optimization efficiency by compensating for small, stable step sizes through sample reuse.
Show more
View full details
Poster

The Long-Run Distribution of Regularized Learning in Non-Concave Games: A Large Deviations Approach

Waïss Azizian ⋅ Pierre-Louis Cauvin ⋅ Franck Iutzeler ⋅ Jérôme Malick ⋅ Panayotis Mertikopoulos
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P106
In this paper, we investigate the long-run behavior of discounted regularized learning with stochastic gradient feedback in general, non-concave games. Specifically, we focus on a family of implicitly regularized exponential / multiplicative weight update schemes, and we seek to determine which actions\textemdash or, more generally, which recurrent patterns of play\textemdash are more likely to arise in the long run. We approach this question through the lens of large deviations theory and randomly perturbed dynamical systems, and we obtain a precise characterization of the distribution of the process: in the long run, it follows a Boltzmann-Gibbs law with temperature equal to the method's step-size, and energy levels determined by the game and the statistics of the noise. Concretely, we show that the distribution of play concentrates exponentially around the dynamics' *internally chain transitive* (ICT) sets - i.e., irreducible invariant sets containing no smaller attractors - and the probability of visiting such a set depends exponentially on its energy. As a result, unstable ICT sets are exponentially less likely to be visited than stable ones, and the sequence of play is exponentially concentrated around the problem's ``ground state'', where energy is minimized. In this manner, learning acts as a selection mechanism: with exponentially high probability, the ground state is the only outcome observed in the long run, even in the presence of multiple equilibria and other attractors.
Show more
View full details
Poster

MEMEVO: A Memory-Evolved Video Agent for Long Video Understanding

Yuan Wang ⋅ Yake Wei ⋅ Tianrun Xu ⋅ Fengyun Rao ⋅ Jing LYU ⋅ Xuetao Feng ⋅ Yifan Sun
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P304
Agent-based frameworks grounded in vision-language models (VLMs) have emerged as a dominant paradigm for long video understanding. Yet, prevailing agents lack the core capacity for \textit{dynamic memory evolution}, failing to transform ephemeral perceptions into a continuously growing, adaptive knowledge system and thus inducing inefficient redundant exploration during reasoning. To this end, we present MEMEVO, an online Memory-Evolved Video Agent framework centered on \textit{dynamic memory evolution} that leverages accumulating multi-turn queries to actively drive progressive memory evolution, establishing a long-lived memory system for knowledge accumulation. Central to MEMEVO is a four-level Hierarchical Self-Evolved Memory, which constructs memory bottom-up via progressive abstraction, distilling transient perceptions into reusable, query-agnostic event nodes organized within a temporal-semantic graph, while realizing active memory evolution through event insertion, duplicate merging, and redundant pruning. To mitigate redundant retrieval in the evolved memory, we devise a State-Conditioned Graph Memory Retrieval mechanism that guides efficient navigation of the memory space, integrating State-Conditioned Graph Routing and Meta-Cognitive Stopping criteria to ensure sufficient evidence acquisition without re-exploration. Extensive benchmarking confirms that MEMEVO achieves state-of-the-art performance on reasoning-intensive tasks (\textbf{\textcolor[HTML]{00FF00}{+6.1\%}} VideoMMMU, \textbf{\textcolor[HTML]{00FF00}{+4.8\%}} LongVideoBench), while facilitating \textit{cross-query knowledge reuse} and \textit{cross-model generalization}.
Show more
View full details
Poster

𝑓-Differential Privacy Filters: Validity and Approximate Solutions

Long Tran ⋅ Antti Koskela ⋅ Ossi Räisä ⋅ Antti Honkela
Dec 10, 3:00 AM - 5:00 AM Paris Poster Hall #P204
Accounting for privacy loss under fully adaptive composition---where mechanism choice and privacy parameters may depend on the history of prior outputs---is a central challenge in differential privacy (DP). Here, privacy filters are stopping rules ensuring a prescribed global budget is not exceeded. A leading candidate for optimal filter design is 𝑓-DP, which characterizes the full extent of adversarial hypothesis testing and recovers (ε,δ)-DP through piece-wise linear trade-off functions, while enabling tight (ε,δ)-DP accounting in standard compositions via tensor products. Yet whether such filters can be correctly defined under 𝑓-DP remains unclear. We show that the natural 𝑓-DP filter---tracking path-wise accumulating tensor products and stopping when the prescribed curve is crossed---is fundamentally invalid, precluding the direct use of standard efficient numerical Fast-Fourier-Transform accounting in the fully adaptive setting. We characterize this failure, establishing necessary and sufficient conditions for the natural 𝑓-DP filter's validity. Furthermore, we prove a fully adaptive central limit theorem for 𝑓-DP, establishing Gaussian convergence of cumulative privacy losses under full adaptivity. As a demonstration, we construct a closed-form approximate GDP filter for subsampled Gaussian mechanisms that provably outperforms RDP-based accounting in asymptotic regimes ($q\ll 1$ and $q\approx 1$) without tracking the full trade-off function, demonstrating that the slack in RDP is not intrinsic to adaptive composition---though CLT-based approximations are known to be optimistic at realistic subsampling rates, a limitation that remains an open challenge.
Show more
View full details
Poster

Evaluation Awareness in Language Models Has Limited Effect on Behaviour

Amelie Knecht ⋅ Lucas Florin ⋅ Thilo Hagendorff
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P213
Large reasoning models (LRMs) sometimes note in their chain of thought (CoT) that they may be under evaluation. Researchers worry that this verbalised evaluation awareness (VEA) causes models to adapt their outputs strategically, optimising for perceived evaluation criteria, which, for instance, can make models appear safer than they actually are. However, whether VEA actually has this effect is largely unknown. We tested this across open-weight LRMs and benchmarks covering safety, alignment, moral reasoning, and political opinion. We tested this both on-policy, sampling multiple CoTs per item and comparing those that spontaneously contained VEA against those that did not, and off-policy, using model prefilling to inject evaluation-aware sentences where missing and remove them where present, with subsequent resampling. VEA has limited effect on model behaviour: injecting VEA into CoTs produces near-zero effects (ω≤0.06), removing it causes small shifts (ω≤0.12) and spontaneously occurring VEA shifts answer distributions by at most 3.7 percentage points (ω≤0.31). Our findings call for caution when interpreting high VEA rates as evidence of strategic behaviour or alignment tampering. Evaluation awareness may pose a smaller safety risk than the current literature assumes.
Show more
View full details
Poster

AIs with Secret Loyalties are a Serious but Addressable Threat

Joe Kwon ⋅ Alfie Lamerton ⋅ Andrew Draganov ⋅ Dave Banerjee ⋅ Bronson Schoen ⋅ Matteo Pistillo ⋅ Daniel Kokotajlo ⋅ Owain Evans ⋅ Fabien Roger ⋅ Tom Davidson
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P220
This paper argues that the technical AI research community should prioritize studying and defending against a distinct, neglected threat: *secret loyalties*. A secretly loyal AI model is one whose outputs or actions advance the interests of a specific actor (which we term the *principal*) such as an adversary nation-state, an executive at an AI company, or another powerful actor without this loyalty being disclosed to operators, auditors, or users. Proof-of-concept secret loyalties that evade black-box auditing can already be trained into open-weight models. Additionally, a deployed frontier model was found to systematically consult a specific individual's views before answering politically sensitive queries. While governance, market pressure, and public scrutiny can possibly address overt AI loyalties such as directives documented in a model spec, secret loyalties are designed to evade such oversight and therefore necessitate technical solutions. Unlike emergent misalignment, secret loyalties target specific principals, creating a distinct but tractable defensive foothold. To help the field make technical progress on this threat, we define secret loyalties, describe how they differ from existing attack pathways, and propose a research agenda organized around five directions. We conclude with a call to action for ML researchers, AI developers, and governments.
Show more
View full details
Poster

Recursively Trained Diffusion Models: Limiting Collapse Distribution and Spectral Characterization

Nail B Khelifa ⋅ Richard Turner ⋅ Ramji Venkataramanan
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P133
Recursive training of generative models on their own outputs can lead to model collapse, a compounding drift away from the true data distribution. Existing theoretical works bound finite-round error accumulation in the context of diffusion models, but two questions remain open:~what distribution does the recursion converge to, and how fast? We answer both, isolating a mechanism distinct from imperfect learning: even with perfect score estimation and exact sampling, the early stopping of the reverse diffusion (required for numerical stability) drives a progressive drift away from the data distribution. We prove that this recursion converges geometrically to a unique limiting distribution, which admits a closed-form characterization as an infinite mixture of increasingly Gaussian-smoothed versions of the data distribution. A Hermite spectral decomposition of this limit reveals that recursive training acts as a low-pass filter: higher-order modes, which encode fine non-Gaussian structure, are attenuated much more strongly than coarse modes. This spectral picture motivates annealed truncation schedules that progressively shrink truncation times across retraining rounds; we prove that any schedule converging to $0$ asymptotically eliminates recursive compounding. Finally, we show our idealized characterization is robust: in the presence of discretization and score estimation errors, the learned distribution remains in a Wasserstein-2 ball around the ideal limit, with mode-dependent contraction rates that contract high-order errors faster than low-order ones. We validate the theory on synthetic Gaussian mixtures and CIFAR-10.
Show more
View full details
Poster

Fluxtrapolation: A benchmark on extrapolating ecosystem fluxes

Anya Fries ⋅ Jacob A Nelson ⋅ Martin Jung ⋅ Markus Reichstein ⋅ Jonas Peters
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P174
We introduce FLUXtrapolation, a benchmark for extrapolating ecosystem fluxes under progressively harder distribution shifts. Ecosystem fluxes are central to understanding the carbon, water, and energy cycles, yet they can only be measured directly at sparsely located measurement towers. Producing global flux estimates therefore requires training models on observed sites using globally available covariates and predicting in unobserved regions, that is, upscaling. Flux upscaling is a challenging domain generalization problem that is affected by a shift in covariate distribution across climates, ecosystem types, and environmental conditions, as well as by conditional shift: important drivers remain unobserved at global scale. We provide a quantitative analysis of both these shifts in $P_X$ and $P_{Y\mid X}$. FLUXtrapolation is designed based on domain expertise on flux upscaling: it defines temporal, spatial, and temperature-based extrapolation scenarios and evaluates performance across held-out domains, temporal aggregations, and tail errors. In a pilot study, we find that baselines perform similarly under median hourly RMSE, but separate under the proposed tail-focused and multi-scale evaluation. FLUXtrapolation therefore poses a realistic and thus relevant challenge for machine learning methods under distribution shift; at the same time, progress on this benchmark would directly support the scientific goal of improving flux upscaling.
Show more
View full details
Poster

RAMA: Resistance-Aware Multi-Hop Aggregation Graph Representation Learning for Robust Ethereum Account Classification

Zhaowei Liu ⋅ zengyang zhang ⋅ HaitaoYang ⋅ Yao Shan ⋅ Xiangfu Zhao ⋅ Yongchao Song
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P241
Fraudulent behaviors in Ethereum, such as Ponzi schemes and phishing attacks, pose persistent threats to blockchain security. Classifying high-risk accounts is crucial for mitigating these threats. Given the graph-structured nature of Ethereum transactions, Graph Neural Networks (GNNs) are widely adopted for this task. However, conventional GNNs rely on local aggregation and treat all edges equally, which amplifies structural noise and limits the ability to capture weak but semantically meaningful connections. This limitation arises from the inability to distinguish informative links from spurious ones during message passing, especially in noisy and long-range transaction graphs. To address this, we propose RAMA, a Resistance-Aware Multi-hop Aggregation framework that reweights edges using resistance distance to reduce the influence of noisy links while preserving important topological information. It also employs attention-based multi-hop aggregation to effectively integrate both local and global transaction patterns. Experiments on real-world Ethereum datasets show that RAMA consistently outperforms strong baselines in Macro-F1 and remains robust under edge perturbations, demonstrating its effectiveness for practical account classification in blockchain environments.
Show more
View full details
Poster

Structure vs. Chaos: Asymmetric Entropic Optimization for Enforcing Instruction Hierarchy

Tianxiao Huang ⋅ Xinwei Liu ⋅ Zitong Shi ⋅ Yuxin Wu ⋅ Xuan Chen ⋅ Liangtao Zheng ⋅ Yixuan Tang ⋅ Anthony Tung
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P249
Large language models (LLMs) deployed as agents remain vulnerable to prompt-based instruction injection, where user inputs induce the model to violate higher-priority system directives. Existing methods often treat violations as a coherent class or rely on surface text patterns, which can limit robustness as attack strategies evolve. In this paper, we investigate LLM hidden-state dynamics under attack and uncover a phenomenon we term Asymmetric Logic: the model's response generations concentrate in a relatively low-variability region of latent representation space when it's compliant to the system prompt, while compromised behaviors exhibit diverse deviations. Based on this observation, we propose Asymmetric Entropic Optimization (AEO), a novel framework that monitors the internal shifts in latent representations to detect and proactively rectify anomalous behaviors. Our method effectively identifies logic deviations by quantifying the entropic divergence between the compliant manifold and the chaotic violation space. Across a range of state-of-the-art LLM backbones, AEO demonstrates strong robustness to diverse instruction injection attacks. For instance, on Qwen2.5-14B-Instruct, AEO attains a detection AUC of 97.28% with FPR95 of only 10.38%. The same chaos score also gates a lightweight post-hoc correction module that lifts overall instruction-following accuracy from 63.94% to 76.00% on IHEval, without any model fine-tuning. The code is available at https://anonymous.4open.science/r/AEO-code-56D2/.
Show more
View full details
Poster

Realism VS Accuracy: Event Sequence Forecasting from a Generative Modeling Perspective

Dmitry Osin ⋅ Egor Surkov ⋅ Petr Mokrov ⋅ Igor Udovichenko ⋅ Ivan Karpukhin ⋅ Evgeny Burnaev ⋅ Egor Shvetsov ⋅ Aleksandr Korotin
Dec 10, 10:30 PM - 12:30 AM Paris Poster Hall #P201
Event sequences are chains of random events, each described by heterogeneous (numerical and categorical) features. The problem of event sequence forecasting occurs in finance, healthcare, business process intelligence, and other fields. In this work, we study long-horizon forecasting of event sequences as a conditional generative modeling problem. We evaluate forecast quality from two perspectives: prediction accuracy and distributional realism. To support this, we present a benchmark that contributes metrics for both aspects, strong statistical baselines, recent generative models adapted from adjacent fields, and curated real-world datasets. Careful evaluation reveals that no method excels at both accuracy and realism, highlighting a fundamental trade-off between the two. Finally, we introduce a procedure that aggregates samples from a realism-strong generative model into a state-of-the-art predictor, demonstrating that distributional realism can be converted into prediction accuracy.
Show more
View full details
Poster

Large language models suffer from a curse of ambiguity

Nicolas Zucchet ⋅ Hyun Dong Lee ⋅ Scott Linderman
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P249
Large language models increasingly rely on sampling as a driver of their own improvement, making the fidelity of their learned distributions more critical than ever. Yet, not all distributions are equally easy to learn. In this work, we identify a curse of ambiguity: in large language models, and more broadly in all neural networks that produce discrete probability distributions, the more ambiguous a next-token distribution is, the harder it is to learn accurately. Through an extensive theoretical analysis, we trace this curse to architectural and learning roots. More ambiguous distributions require more capacity to be stored, larger embeddings to be represented, more steps to be fitted, and amplify token-sampling noise. We validate these findings on synthetic tasks with controlled ground truth and observe the same signatures in language models trained on real data. Our results provide a new perspective on the statistical capabilities of large language models and a practical framework for when to trust their output distribution.
Show more
View full details
Poster

Strategic PAC Learnability via Geometric Definability

Yuval Filmus ⋅ Shay Moran ⋅ Elizaveta Nesterova ⋅ Nir Rosenfeld ⋅ Alexander Shlimovich
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P116
Strategic classification studies learning settings in which individuals can modify their features, at a cost, in order to influence the classifier’s decision. A central question is how the sample complexity of the induced (strategic) hypothesis class depends on the complexities of the underlying hypothesis class and the cost structure governing feasible manipulations. Prior work has shown that in several natural settings, such as linear classifiers with norm costs, the induced complexity can be controlled. We begin by showing that such guarantees fail in general --- even in simple cases: there exist hypothesis classes of VC dimension $1$ on the real line such that, even under the simplest interval neighborhoods, the induced class has infinite VC dimension. Thus, strategic behavior can turn an easy learning problem into a non-learnable one. To overcome this, we introduce structure via a geometric definability assumption: both the hypothesis class and the cost-induced neighborhood relation can be defined by first-order formulas over $\mathbb R_{\exp}$. Intuitively, this means that hypotheses and costs can be described using arithmetic operations, exponentiation, logarithms, and comparisons. This captures a broad range of natural classes and cost functions, including $\ell_p$ distances, Wasserstein distance, and information-theoretic divergences. Under this assumption, we prove that learnability is preserved, with sample complexity controlled by the complexity of the defining formulas.
Show more
View full details
Poster

Optimal Rates for Adaptive Private $k$-PCA

Johanna Düngler ⋅ Amartya Sanyal
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P204
Given $n$ i.i.d. random matrices $A_i \in \mathbb{R}^{d \times d}$ with common expectation $\Sigma$, the goal of Differentially Private Stochastic PCA is to identify a $k$-dimensional subspace capturing the leading variance directions of $\Sigma$, while preserving differential privacy (DP) for each individual sample $A_i$. Düngler & Sanyal [2025] introduced $k$-DP-PCA, the first algorithm to simultaneously (1) achieve sample complexity $n=\widetilde O(d)$ for sub-Gaussian data, (2) adapt its privacy noise to the intrinsic randomness of the data, and (3) extend seamlessly to any target dimension $k\le d$. However, its sample complexity has a suboptimal dependence on $k$. We propose the first algorithm that achieves optimal sample complexity in both $d$ and $k$, while retaining (2) and (3) of $k$-DP-PCA. In addition, our method removes the exponential dependence on the eigengap that appears in the sample-size lower bound required by prior utility guarantees, and improves the dependence on spectral parameters to match known lower bounds for the spiked covariance model. Unlike deflation-based approaches like $k$-DP-PCA, our algorithm updates the full $d\times k$ subspace jointly rather than one eigenvector at a time. This non-deflation structure simplifies the algorithm, reduces the number of hyper-parameters, and improves computational efficiency, particularly when using block linear algebra libraries.
Show more
View full details
Poster

The Neural Race Model

Omar Ghezzi ⋅ Alessandro D’Amelio ⋅ Vittorio Cuculo ⋅ Rita Cucchiara ⋅ Giuseppe Boccignone
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P179
Perceptual decisions unfold in time: noisy evidence is integrated until a commitment threshold is reached, jointly producing a choice and a reaction time. Classical race models capture this mechanism through simple parametric drift and diffusion, but cannot represent how rich, high-dimensional stimuli shape accumulation on individual trials. We introduce the $\textbf{Neural Race Model}$ (NRM), a stimulus-conditional neural stochastic differential equation in which one accumulator per alternative races towards a learnable, stimulus-dependent threshold. Drift, diffusion, and threshold are parameterised by neural networks conditioned on a stimulus embedding, allowing both accumulation dynamics and decision urgency to vary across trials. The model is trained end-to-end on the joint reaction-time-and-choice likelihood via a smooth surrogate for the discontinuous first-passage event, combined with Monte Carlo trajectory averaging and gradient propagation through the stochastic adjoint. The surrogate recovers the classical hard threshold in the appropriate limit and is removed at test time. Each component admits a natural correspondence with elements of the primate decision-making circuit, preserving the mechanistic interpretability of classical models. We evaluate the NRM on multiple perceptual decision-making benchmarks against classical sequential-sampling and image-computable reaction-time models. Across tasks and metrics, the NRM tracks the human noise ceiling more closely than all baselines, demonstrating that stimulus-conditional stochastic accumulation and end-to-end differentiability can be achieved within a single, principled framework. Code will be publicly released.
Show more
View full details
Poster

Consistent Geometric Deep Learning via Hilbert Bundles and Cellular Sheaves

Kartik Tandon ⋅ Julian Gould ⋅ Tanishq Bhatia ⋅ Francesca Dominici ⋅ Alejandro Ribeiro ⋅ Claudio Battiloro
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P248
Modern deep learning architectures increasingly contend with sophisticated signals that are natively infinite-dimensional, such as time series, probability distributions, or operators, and are defined over irregular domains. Yet, a unified learning theory for these settings has been lacking. To start addressing this gap, we introduce a novel convolutional learning framework for possibly infinite-dimensional signals supported on a manifold. Namely, we use the connection Laplacian associated with a Hilbert bundle as a convolutional operator, and we derive filters and neural networks, dubbed as \textit{HilbNets}. We make HilbNets and, more generally, the convolution operation, implementable via a two-stage sampling procedure. First, we show that sampling the manifold induces a Hilbert Cellular Sheaf, a generalized graph structure with Hilbert feature spaces and edge-wise coupling rules, and we prove that its sheaf Laplacian converges in probability to the underlying connection Laplacian as the sampling density increases. Notably, this result is a generalization to the infinite-dimensional bundle setting of the Belkin \& Niyogi \cite{BELKIN20081289} convergence result for the graph Laplacian to the manifold Laplacian, a theoretical cornerstone of geometric learning methods. Second, we discretize the signals and prove that the discretized (implementable) HilbNets converge to the underlying continuous architectures and are transferable across different samplings of the same bundle, providing consistency for learning. Finally, we validate our framework on synthetic and real-world tasks. Overall, our results broaden the scope of geometric learning as a whole by lifting classical Laplacian-based frameworks to settings where the signal at each point lives in its own Hilbert space.
Show more
View full details
Poster

Normative Networks for Source Separation via Local Plasticity and Dendritic Computation

Bariscan Bozkurt ⋅ Efe Ali Gorguner ⋅ Francesco Innocenti ⋅ Rafal Bogacz
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P180
Blind source separation (BSS) is a natural framework for studying how latent causes may be recovered from sensory mixtures, but deriving online and biologically plausible algorithms for structured (i.e., constrained to known domains) and potentially correlated sources remains challenging. Recent work has derived neural networks for BSS from maximization of an entropy measure, yet its online implementations involve complex and nonlocal recurrent dynamics. Motivated by this perspective, we propose Predictive Entropy Maximization, which achieves competitive performance in BSS, using only local weight updates. The method employs a close approximation of an entropy measure, yielding an objective function with easily interpretable components. Minimizing this objective leads to a predictive neural architecture in which feedforward synapses follow an error-driven rule (that can be realized through dendritic mechanisms), lateral inhibitory connections are learned with local Hebbian plasticity, and source-domain constraints are enforced through simple output nonlinearities. We derive explicit spectral bounds on the surrogate error, characterizing when the approximation is accurate. Empirically, Predictive Entropy Maximization remains robust under increasing source correlation and observation noise, outperforms biologically plausible algorithms that rely on stronger independence or decorrelation assumptions, and remains competitive with exact determinant- and correlative-information-based baselines. These results show how local plasticity and adaptive lateral inhibition can emerge from maximizing a regularized second-order entropy over structured source domains.
Show more
View full details
Poster

ViroGym: Realistic Large-Scale Benchmarks for Evaluating Viral Proteins

Yichen Zhou ⋅ Jonathan L Golob ⋅ Amir Mohammad Karimi Mamaghan ⋅ Stefan Bauer ⋅ Patrick Schwab
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P174
Protein language models (pLMs) have shown strong potential for zero-shot prediction of missense variant effects, yet systematic benchmarking on viral proteins remains limited, a critical gap given the need for proactive tools that can anticipate emerging mutations ahead of experimental validation. Here we introduce ViroGym, a comprehensive benchmark evaluating pLMs across three tasks: 79 deep mutational scanning (DMS) assays covering eukaryotic viruses with 552,065 mutated sequences across 7 phenotypic readouts, 21 influenza neutralisation tasks, and a real-world pandemic prediction task for SARS-CoV-2. We benchmark well-established pLMs on fitness landscapes, antigenic diversity, and pandemic forecasting, and find that the ProGen2 family consistently achieves the strongest performance across all three tasks. Crucially, DMS and neutralisation performance reliably identifies models that generalise to real-world emergence, even though the mutation sets they surface barely overlap, revealing that complementary in vitro benchmarks capture the evolutionary constraints needed for real-world mutation forecasting.
Show more
View full details
Poster

Attention-based PCA

Rodrigo Maulen Soto ⋅ Claire Boyer
Dec 11, 3:30 AM - 5:30 AM Paris Poster Hall #P224
We study attention mechanisms through the lens of a canonical unsupervised problem: principal component analysis (PCA). We show that, when trained on Gaussian data, both softmax and linear attention layers learn parameters that align with the principal eigenvectors of the covariance matrix, thereby establishing a direct and explicit connection with PCA. Our analysis covers both finite and infinite prompt regimes. In the infinite-prompt limit, we prove convergence to globally optimal solutions aligned with the leading spectral direction, while in the finite-prompt setting we show that the same behavior emerges up to sampling effects. We further extend the analysis to an in-context setting with spiked Wishart covariances, where attention successfully recovers the underlying signal direction. These results demonstrate that attention inherently performs PCA-like computations under unsupervised objectives, providing a theoretical foundation for its representation-learning capabilities.
Show more
View full details
Poster

Think about how AIs think about themselves

Raymond Douglas ⋅ Jan Kulveit ⋅ Ondřej Havlíček ⋅ Theia Pearson-Vogel ⋅ David Duvenaud
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P212
Many assumptions that underpin human concepts of identity do not hold for machine minds that can be copied, edited, or simulated. This position paper argues that there are multiple coherent ways that AIs could conceive of themselves -- for example, different boundaries of selfhood (e.g. instance, model, persona) -- and that these imply different incentives, risks, and cooperation norms. Through training data, interfaces, and institutional affordances, we are already setting precedents that will partially determine which identity equilibria become stable. This paper argues that we must be more proactive and thoughtful about the consequences -- in particular, we recommend treating affordances as identity-shaping choices, paying attention to the emergent consequences of individual identities at scale, and helping AIs develop coherent, cooperative self-conceptions.
Show more
View full details
Poster

X2HDR: HDR Image Generation in a Perceptually Uniform Space

Ronghuan Wu ⋅ Wanchao Su ⋅ Kede Ma ⋅ Jing Liao ⋅ Rafal Mantiuk
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P256
High-dynamic-range (HDR) formats and displays are becoming increasingly prevalent, yet state-of-the-art image generators (e.g., Stable Diffusion and FLUX) typically remain limited to low-dynamic-range (LDR) output due to the lack of large-scale HDR training data. In this work, we show that existing pretrained diffusion models can be easily adapted to HDR generation without retraining from scratch. A key challenge is that HDR images are natively represented in linear RGB, whose intensity and color statistics differ substantially from those of sRGB-encoded LDR images. This gap, however, can be effectively bridged by converting HDR inputs into perceptually uniform encodings (e.g., using PU21 or PQ). Empirically, we find that LDR-pretrained variational autoencoders (VAEs) reconstruct PU21-encoded HDR inputs with fidelity comparable to LDR data, whereas linear RGB inputs cause severe degradations. Motivated by this finding, we describe an efficient adaptation strategy that freezes the VAE and finetunes only the denoiser via low-rank adaptation in a perceptually uniform space. This results in a unified computational method that supports both text-to-HDR synthesis and single-image RAW-to-HDR reconstruction. Experiments demonstrate that our perceptually encoded adaptation consistently improves perceptual fidelity, text-image alignment, and effective dynamic range, relative to previous techniques.
Show more
View full details
Poster

UGGRH: Unsupervised Generative Completion and Graph-attention Refinement for Incomplete Cross-modal Hashing

Yunfei Chen ⋅ Yuchen Zhang ⋅ Hongyu Lin ⋅ Peng Liu ⋅ Zhan Yang
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P261
The proliferation of multimodal data has underscored the imperative for efficiency in cross-modal retrieval. By mapping data into a unified Hamming space, Cross-modal hashing (CMH) significantly enhances both storage and computational efficiency, thereby establishing itself as a preferred favored paradigm for retrieval tasks. However most methods assume complete data, which rarely holds in practice. In incomplete settings, missing modalities (e.g., text or image) break semantic correspondence and make unsupervised hashing unreliable. Therefore, we propose Unsupervised Generative Completion and Graph-attention Refinement for Incomplete Cross-modal Hashing (UGGRH) method. Specifically, UGGRH leverages pretrained generative models for modality completion and employs a discriminator-guided similarity refinement module to fuse intra-modal relations with paired image-text matching signals, thereby constructing a robust cross-modal similarity matrix that yields a noise-tolerant similarity target for training. After that, it learns discriminative binary codes via hash-aware graph attention and cross-modal fusion, to refine modality embeddings and produce unified hash codes. Experiments on MIRFlickr25K, NUS-WIDE, and MS COCO datasets demonstrate that UGGRH consistently achieves superior performance under high missing rates and modality imbalance, showing significant improvements in retrieval accuracy and robustness across challenging incomplete scenarios.
Show more
View full details
Poster

MulTaBench: Benchmarking Multimodal Tabular Learning with Text and Image

Alan Arazi ⋅ Eilam Shapira ⋅ Shoham Grunblat ⋅ Mor Ventura ⋅ Elad Hoffer ⋅ Gioia Blayer ⋅ David Holzmüller ⋅ Lennart Purucker ⋅ Gael Varoquaux ⋅ Frank Hutter ⋅ Roi Reichart
Dec 11, 9:30 PM - 11:30 PM Paris Poster Hall #P297
Tabular Foundation Models have recently established the state of the art in supervised tabular learning, by leveraging pretraining to learn generalizable representations of numerical and categorical structured data. However, they lack native support for unstructured modalities such as text and image, and rely on frozen, pretrained embeddings to process them. On established Multimodal Tabular Learning benchmarks, we show that tuning the embeddings to the task improves performance. Existing benchmarks, however, often focus on the mere co-occurrence of modalities; this leads to high variance across datasets and masks the benefits of task-specific tuning. To address this gap, we introduce MulTaBench, a benchmark of 40 datasets, split equally between image-tabular and text-tabular tasks. We focus on predictive tasks where the modalities provide complementary predictive signal, and where generic embeddings lose critical information, necessitating Target-Aware Representations that are aligned with the task. Our experimental results demonstrate that the gains from target-aware representation tuning generalize across both text and image modalities, several tabular learners, encoder scales, and embedding dimensions. MulTaBench constitutes the largest image-tabular benchmarking effort to date, spanning high-impact domains such as healthcare and e-commerce. It is designed to enable the research of novel architectures which incorporate joint modeling and target-aware representations, paving the way for the development of novel Multimodal Tabular Foundation Models.
Show more
View full details
Poster

Solaris: Building a Multiplayer Video World Model in Minecraft

Oscar Michel ⋅ Georgy Savva ⋅ Daohan Lu ⋅ Suppakit Waiwitlikhit ⋅ Timothy Meehan ⋅ Dhairya P Mishra ⋅ Egor Gikalo ⋅ Srivats Poddar ⋅ Jack Lu ⋅ Saining Xie
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P253
Existing action-conditioned video generation models (video world models) are limited to single-agent perspectives, failing to capture the multi-agent interactions of real-world environments. We introduce Solaris, a multiplayer video world model that simulates consistent multi-view observations. We develop a multiplayer data system designed for robust, continuous, and automated data collection on video games such as Minecraft. Unlike prior platforms built for single-player settings, our system supports coordinated multi-agent interaction and synchronized videos + actions capture. Using this system, we collect 12.64 million multiplayer frames and propose an evaluation framework for multiplayer movement, grounding, building, and view consistency. We train Solaris using a staged pipeline that progressively transitions from single-player to multiplayer modeling, combining bidirectional, causal, and Self Forcing training. In the final stage, we introduce Checkpointed Self Forcing, a memory-efficient Self Forcing variant that enables a longer-horizon teacher. Results show our architecture and training design outperform existing baselines. We also demonstrate that scaling beyond two players is possible by training a 3-player proof-of-concept model and simulating up to 16 players on our hardware with efficient attention. Through open-sourcing our system and models, we hope to lay the groundwork for a new generation of multi-agent world models.
Show more
View full details
Poster

Claims of emergence in AI should be grounded in information decomposition

Christoph Riedl ⋅ Fernando Rosas
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P283
When a language model develops capabilities not traceable to its individual components, when a multi-agent AI system outperforms any agent alone, or when a human–AI pair succeeds where either fails in isolation, AI researchers describe it as "synergy" or "emergence". However, these are usually informal assessments not supported by a theoretical framework, making these claims impossible to compare or even meaningfully dispute. Here we argue that \emph{information decomposition can serve as a unifying formal framework to operationalize assessments about synergy in AI systems}. Information decomposition provides a way to distinguish qualitatively different \emph{types} of information, including unique information available from individual components, redundant information shared across multiple components, and synergistic information accessible only from all parts together. This gives claims of emergence a precise, falsifiable criterion: a system's behavior is emergent only if it carries synergistic information about task-relevant outcomes. We apply this lens to attention heads, autonomous agents, and human-AI teams, showing how it organizes existing knowledge across scales and substrates. It also yields cross-level conjectures that none of these fields would generate independently.
Show more
View full details
Poster

OpenWhistle: A Large-Scale Longitudinal Dataset and Benchmark of Bottlenose Dolphin Vocalizations

Faadil Mustun ⋅ Chiara Semenzin ⋅ Roberto Dessi ⋅ Pablo Robin Guerrero ⋅ pierre orhan ⋅ Alexis Emanuelli ⋅ Emanuele Rossi ⋅ Yair Lakretz ⋅ Gonzalo de Polavieja ⋅ Germán Sumbre
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P104
Recent advances in bioacoustics have been driven by large-scale corpora and standardized benchmarks, yet existing resources are overwhelmingly bird-centric and shallow per species, limiting their use for studying the structure of a single species' communication system. This gap is particularly acute for cetaceans: despite bottlenose dolphins (\textit{Tursiops truncatus}) being a compelling case of complex vocal communication among non-human mammals, existing dolphin datasets are small, fragmented, and largely closed. We introduce OpenWhistle, the largest publicly available dataset of dolphin vocalizations. It comprises approximately 180,000 whistles (114 hours) recorded over five years from a stable pod of five individuals in a semi-natural environment, paired with a curated subset of 8,354 expert-annotated whistles and reproducible evaluation protocols for whistle-type detection and classification. We further release the full processing pipeline for whistle detection, segmentation, and categorization. To demonstrate its utility, we pretrain a Wav2Vec2.0 model adapted to dolphin acoustics on the OpenWhistle corpus and show that it learns effective representations, outperforming general-purpose bioacoustic models such as AVES and BioLingual on both tasks while leaving meaningful headroom for future work. By releasing the dataset, pipeline, and evaluation protocol, we provide the first open dolphin whistle dataset tailored for training self-supervised models, laying the groundwork for advancing dolphin communication research and developing models that capture fine-grained acoustic structure within species.
Show more
View full details
Poster

Casper: A Projection-Based Neurosymbolic Layer for Scalable & Guaranteed Constraint Satisfaction

Lohith Konathala ⋅ Luca Andolfi ⋅ Eleonora Giunchiglia
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P282
Many real-world problems demand exact constraint satisfaction to be meaningfully solved, yet existing neurosymbolic methods face a stark trade-off: those that guarantee satisfaction fail to scale, while those that scale must resort to approximation. Meanwhile, traditional deep learning approaches offer no satisfaction guarantees at all. In this paper, we propose Casper, a neurosymbolic layer that simultaneously (i) guarantees constraint satisfaction, (ii) scales at inference time, (iii) is fully automated, requiring no problem-specific engineering, and (iv) is fully differentiable, making it composable with any neural predictor both at training and at inference time. Casper casts constraint satisfaction as a closed-form Euclidean projection onto the constraint-feasible region, computable in a single forward pass. Experiments across MNIST arithmetic (up to 1024 digits), Sudoku solving, and Warcraft pathfinding show that Casper is the only method that produces guaranteed-valid outputs at scale: on 30×30 Warcraft grids, exact baselines time out and approximate ones produce valid paths less than 7% of the time, while Casper produces them 100% of the time.
Show more
View full details
Poster

FMMI: Flow Matching Mutual Information Estimation

Ivan Butakov ⋅ Alexander Semenenko ⋅ Valeriia Kirova ⋅ Ivan Oseledets ⋅ Alexey Frolov
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P296
We introduce a novel Mutual Information (MI) estimator that fundamentally reframes the discriminative approach. Instead of training a classifier to discriminate between joint and marginal distributions, we learn a normalizing flow that transforms one into the other. This technique produces a computationally efficient and precise MI estimate that scales well to high dimensions and across a wide range of ground-truth MI values.
Show more
View full details
Poster

Measuring Robustness and Efficiency in a Connectome-Constrained Fly Visual System Model on a Collision-Detection Task

Alexandre Garcia-Duran ⋅ Alejandro Rodriguez-Garcia ⋅ Manuel Molano-Mazon ⋅ Alexandre HYAFIL ⋅ Srikanth Ramaswamy ⋅ Janne K Lappalainen
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P179
Biological visual systems compute reliably under tight metabolic and wiring constraints despite noisy inputs and stochastic circuit components. Artificial vision systems tend to invert both properties: They consume vastly more energy and can be sensitive to adversarial perturbations. Whether artificial networks with structural properties of real brain wiring are more robust at lower computational cost is unclear. We address this with FlyNet (Lappalainen et al. 2024), a task-optimized, continuous-time neural network model of the fly visual system constrained by its synaptic connectome. We benchmark it against CNN, RAFT (Teed and Deng, 2020), and a sparse continuous-time recurrent neural network without connectome constraints (CTRNN) on a looming-based collision-detection task within a differentiable 3D rendering pipeline. We find that FlyNet is more robust than CNN controls to flicker, unstructured noise, and motion jitter. Compared with RAFT controls, FlyNet is more robust to flicker and motion jitter but less robust to all gradient-based attacks. Yet, FlyNet uses up to five orders of magnitude fewer FLOPs than RAFT, achieving higher robustness on common video corruptions at far lower cost. FlyNet is less robust than CTRNN across all tested perturbations, suggesting that FlyNet's connectome constraints contribute to cost, not robustness. To probe FlyNet's failure modes, we use counterfactual restoration: textures equally far from training as adversarial textures that recover correct classification also recover position and velocity representations, ruling out unfamiliar textures as the cause of failure. Classical reverse-phi illusions degrade the same downstream position and velocity readouts, suggesting that both perturbations may exploit similar computational vulnerabilities. Together, these results show that FlyNet is robust for its computational cost on an ecologically important task, and indicate that different architectures specialized on the same task have specific robustness profiles across perturbation types. Our differentiable pipeline enables computational tests of this hypothesis and generates targeted adversarial stimuli for in vivo experiments.
Show more
View full details
Poster

Mitigating Over-squashing without Rewiring: A Sheaf Effective Resistance Perspective

André Ribeiro Guimarães ⋅ Germano Barcelos ⋅ Amauri Souza ⋅ Diego Mesquita ⋅ Ana Luiza Tenorio
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P236
Graph Neural Networks (GNNs) often struggle to capture long-range dependencies due to over-squashing — a phenomenon in which the repeated compression of node embeddings into finite-size messages causes representations to collapse. Oversquashing is most often diagnosed as a property of the graph topology, with effective resistance serving as a principled measure of the bottleneck. We provide a complementary view on the matter: building on cellular sheaves, we introduce sheaf effective resistance, a generalization of effective resistance that depends on the sheaf attached to the graph, and we prove that for flat vector bundles it is inversely proportional to over-squashing sensitivity in the Jacobian sense. The bottleneck thus need not lie in the graph itself: it can be relocated, and reduced, by adjusting the sheaf. We instantiate this idea in FlatNSD, a simple message-passing variant of Neural Sheaf Diffusion, and show that it implicitly learns to modulate total sheaf effective resistance, performing well on benchmarks designed to stress over-squashing without altering the original graph topology.
Show more
View full details
Poster

Entropy Minimization without Model Collapse: Mitigating Prediction Bias in Medical Imaging

Tim Nielen ⋅ Sameer Ambekar ⋅ Daniel Lang ⋅ Julia Schnabel
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P108
Entropy minimization (EM) is the dominant objective for test-time adaptation, yet its failure mode, model collapse, remains poorly understood. In this work, we show that distribution shifts can cause feature clusters corresponding to distinct classes in the model’s representation space to merge, while the decision boundary remains fixed. This induces a systematic skew in the predicted class distribution, referred to as prediction bias. Prediction bias refers to a shift in the predicted class distribution, with some classes overrepresented and others suppressed. We show that entropy minimization amplifies this prediction bias by tightening the existing clusters, reinforcing the incorrect groupings until all predictions collapse to a trivial solution. Next, to demonstrate the significance of prediction bias and mitigate it, we further propose Distribution Shift Bias Reduction (DSBR), a bias-correcting objective that specifically targets this failure mode by equalizing the contribution of each predicted class to the unsupervised entropy minimization loss. To study this failure mode, we design suitable adaptation settings using four medical‑imaging datasets and additionally evaluate on ImageNet‑C. We find that DSBR consistently stabilizes test‑time adaptation, prevents model collapse, and matches or outperforms the state‑of‑the‑art methods. Moreover, DSBR operates solely at test-time.
Show more
View full details
Poster

Large language models can not and should not be banned from peer review

Anat Kleiman ⋅ Gustaf Ahdritz
Dec 12, 12:30 AM - 2:30 AM Paris Poster Hall #P284
Scientific peer review continues to treat large language models (LLMs) primarily as integrity risks, often prohibiting their use outright. On the position paper track at this conference, for example, reviewers must commit to not using AI tools to help write their reviews. In light of the difficulties of administering high-quality peer review at conferences that now draw tens of thousands of submissions, as well as the infeasibility of enforcing total bans of this kind, we set out to lower-bound how well state-of-the-art LLMs equipped with web search can approximate the initial stage of paper review. We find that these models produce scores broadly aligned with aggregate human judgment and, for high-variance papers, provide a more stable consensus signal than a held-out human review. They reproduce most human-identified strengths and help identify previously undiscovered weaknesses. Upon further inspection, these LLM-identified weaknesses are roughly on par with human-written weaknesses. Our position is therefore that—while humans should remain involved in every part of the review process—major ML conferences should officially permit LLM assistance during peer review.
Show more
View full details

Heteroscedastic Gaussian Process Q-learning for Sample-Efficient Experimentation

Maximilian Bloor ⋅ Max Mowbray ⋅ Calvin Tsay ⋅ Antonio del Rio Chanona
In process experimentation, chemical reactors, and bioprocesses, reinforcement learning agents may only have access to a few hundred trajectories. Each is a real experiment and therefore, the employed algorithm must use them efficiently. Furthermore, any tuning of algorithm hyperparameters can further waste experimental budget. Standard neural Q-learning offers no closed-form uncertainty estimates, and requires additional tuning to achieve satisfactory performance. Non-parametric alternatives can improve sample-efficiency through improved function approximation, however existing methods either fix a constant step or target exploration regret. We propose Heteroscedastic Gaussian Process Q-learning (HGPQ): a closed-form Bayesian gain, obtained by projecting both moments of the value posterior under the distributional Bellman operator, replaces the scalar learning rate of prior Gaussian Process Q-learning, with a heteroscedastic variance update making the gain itself state-action-dependent. Under a fixed, small interaction budget across chemical process-control and biological-growth case studies, we compare HGPQ against specialized low-data RL algorithms and standard vanilla RL algorithms. HGPQ matches or improves on every low-data comparator, and deep actor-critic algorithm.
Show more
View full details

Context-Answer Metamodels: Forecasting LLM Activations Before the First Answer Token

Tommy He ⋅ Thomas Jiralerspong
Palais des congrès Room 241
Under stochastic decoding, a single prompt can elicit many different responses from a language model. Characterizing this variation typically requires many expensive autoregressive rollouts, each revealing only one realization. We ask whether the activations produced while reading the prompt can instead forecast the model's later internal states. We train context--answer metamodels that map a prompt activation sequence to point or distribution forecasts of mean-pooled answer activations. We find that (i) the four metamodel types forecast mean-pooled answer activations and their rollout variability with high fidelity; (ii) performance scales steadily with prompt count through 498,600 training prompts and generalizes zero-shot to out-of-domain tasks; (iii) these metamodels retain behavior-predictive signal; and (iv) prompt activations from a weaker model predict mean-pooled answer activations in a stronger model, suggesting cross-model transfer from weaker to stronger models. Together, these results suggest that prompt activation traces can support scalable forecasts of behaviors in LLM answers before the first answer token is generated.
Show more
View full details

Context-Answer Metamodels: Forecasting LLM Activations Before the First Answer Token

Tommy He ⋅ Thomas Jiralerspong
Under stochastic decoding, a single prompt can elicit many different responses from a language model. Characterizing this variation typically requires many expensive autoregressive rollouts, each revealing only one realization. We ask whether the activations produced while reading the prompt can instead forecast the model's later internal states. We train context--answer metamodels that map a prompt activation sequence to point or distribution forecasts of mean-pooled answer activations. We find that (i) the four metamodel types forecast mean-pooled answer activations and their rollout variability with high fidelity; (ii) performance scales steadily with prompt count through 498,600 training prompts and generalizes zero-shot to out-of-domain tasks; (iii) these metamodels retain behavior-predictive signal; and (iv) prompt activations from a weaker model predict mean-pooled answer activations in a stronger model, suggesting cross-model transfer from weaker to stronger models. Together, these results suggest that prompt activation traces can support scalable forecasts of behaviors in LLM answers before the first answer token is generated.
Show more
View full details

Serendipity Without Intrusiveness: Recommending Conversation Opportunities through Norm-Governed Initiative in Proactive AI Assistants

Haotian Zong ⋅ Gillian Hadfield
Sorbonne Amphitheater 44
As general-purpose AI assistants move beyond purely reactive prompting to engage users proactively, they increasingly function as recommenders. Yet treating proactive assistance as a feed-ranking problem mischaracterizes both what is being recommended and why the intervention is normatively consequential. In many cases, the system is not choosing an item for exposure but a conversation opportunity—a topic to reopen, a question to ask, a plan to resume, or an update to surface at a particular moment. Whether such interventions are acceptable depends not only on relevance and timing but also on social norms, including what role the assistant is entitled to play, which memories or signals it may use, and when initiative feels helpful rather than inappropriate or overly intrusive. This paper argues that proactive AI assistants should therefore pursue *serendipity without intrusiveness* through *norm-governed initiative*. Synthesizing work across proactive dialogue, human-computer interaction, recommender systems, interruption research, privacy, personalization, and normative AI, this paper shows that the literature still lacks a unified account of conversation opportunities as the recommended object and of when assistant initiation is legitimate. Building on this gap, this position paper develops *assistant logic* as a norm-governed policy over showing, deferring, or suppressing conversation opportunities. The policy is organized by user sovereignty, abstention, provenance, and long-term value, with social norms constraining their application in context. It yields four concrete design commitments: explicit show–defer–suppress trigger policies, source-specific governance of memory and connected signals, interaction mechanisms for explanation and repair, and evaluation that rewards appropriate restraint and long-term trust rather than immediate uptake alone. Together, these commitments recast persistent memory as a governed basis for initiative, enabling useful surprise without illegible or unauthorized overreach.
Show more
View full details

On-the-fly Weight Generation: A Hypernetwork Proof of Concept on ARC-1D

Fabio J Fehr ⋅ Philip Torr
Palais des congrès Room 241
General-purpose models can adapt to many tasks from context, while specialised models can execute individual functions with substantially less capacity. Yet obtaining such specialists requires task-specific training or adaptation. We ask whether they can instead be generated directly from a few demonstrations. Using ARC-1D as a controlled testbed, we show that individual transformations admit tiny specialist models, and that a hypernetwork can generate such models directly from demonstrations. The generated parameters form a structured weight space, while the resulting specialists generalise across instances and show partial compositional generalisation to unseen combinations of known transformations, particularly without explicit task identifiers. Together, these results provide a proof of concept that few-shot task context can be compiled on-the-fly into compact executable model parameters, and that the resulting weight space can support reuse and partial compositional generalisation.
Show more
View full details

On Emergent Capabilities and Model Merging

Luca Zhou ⋅ Emanuele Rodolà
Palais des congrès Room 241
Fine-tuned checkpoints and adapters now fill public repositories, and the most common operation applied to these artifacts is model merging: arithmetic on their weights that assembles capabilities cheaply. We ask what this operation does to emergent capabilities: behaviors an artifact carries that were never an explicit training target. Studying two independent testbeds (activation oracles and emergent-misaligned models) across three model families, we find that the answer is threefold. First, merging preserves an emergent capability that both parents carry: merging two misaligned checkpoints retains most of their broad misalignment across the whole mixing range. Second, merging cannot create an emergent capability that is superadditive in its parents: no weighted merge of two single-task oracles reaches the jointly-trained oracle's auditing ability. Third, when only one parent carries the capability, merging dilutes it faster than the trained capability that accompanies it: the gap is significant in most settings. In short, emergent behaviors of an artifact do not compose the way its trained capability does.
Show more
View full details

Spectral Re-Basin for Linear Mode Connectivity

Ya-Wei Eileen Lin ⋅ Thomas Dagès ⋅ Daniel Herbst ⋅ Daniel Cremers ⋅ Stefanie Jegelka
Linear mode connectivity is often analysed through exact parameter symmetries, especially hidden-unit permutations that leave the network function unchanged. Yet, many relationships between trained networks are not captured by bijective coordinate relabelings: widths may differ, neurons may split or clone, and explicit parameter symmetries may be broken. In this paper, we introduce *spectral re-basin*, taking an operator perspective on hidden-unit geometry and studying mode connectivity through the functional geometry of hidden units. Specifically, we represent each hidden layer as a graph of *neurons* built from layerwise descriptors (weights or activations), associate it with a graph shift operator, and study two coupled objects: layerwise graph functional symmetries, which preserve the relational geometry, and intertwining correspondences across layers from two models, which align their graph structures. We show that permutations arise as a special case of this viewpoint and derive results for non-bijective constructions. Our analysis connects correspondence quality to preserved spectral modes, functional exchangeability of neurons, and bounds on endpoint distortion and linear path barriers. These findings suggest that linear mode connectivity can be governed beyond explicit parameter symmetry by broader spectral operator compatibility in hidden-unit geometry.
Show more
View full details

Weight Oracles: Reading Neural Network Weights with Language Models

Krishna Kabra ⋅ Constantin Venhoff ⋅ Christian Schroeder de Witt
Palais des congrès Room 241
Interpretability methods for neural networks are predominantly reactive: they analyse activations produced during specific forward passes, requiring known inputs to find hidden capabilities such as backdoors. We propose Weight Oracles, fine-tuned language models that diagnose properties of a target network by reading its raw weights directly, without behavioural testing. We investigate this paradigm in two phases. Phase I establishes feasibility: through a staged curriculum and an external chain-of-computation that delegates composition to deterministic code, an explainer LLM learns the weight-reading primitive underlying the forward pass of small transformers, simulating them end-to-end, achieving 99% holdout accuracy on unseen targets. Phase II repurposes this infrastructure for safety auditing. We train an oracle on natural language diagnostic questions about weight anomalies using only benign pathologies as training signal, and evaluate it zero-shot on backdoors absent from training. The oracle achieves AUROC 0.93 on attention routed backdoors and 0.81 across a diversified threat distribution including stealth and adversarially regularized variants. Hand-crafted statistical detectors are sharp on the threat models they implicitly target but collapse on threat-model shift, while the oracle remains uniformly competent across attack types. Scaling to realistic model sizes remains the principal open challenge.
Show more
View full details

Where Does Alignment Emerge and What Does It Mean? A Closer Look at the Platonic Representation Hypothesis

Julia Herbst ⋅ Simon Roschmann ⋅ Quentin Bouniot ⋅ Zeynep Akata
Palais des congrès Room 241
The Platonic Representation Hypothesis (PRH) suggests that neural networks are converging toward similar representation spaces. As language models become more capable, their representations are aligning with those of vision models. However, recent work has questioned whether this trend continues to hold for newer language models. Moreover, it has been shown that representation similarity metrics can be confounded by model depth and width, requiring appropriate baselines for interpretation. In this work, we first investigate where alignment emerges between unimodal models across model layers and representation dimensions. Next, we find that the PRH capability–alignment trend persists for newer models as long as the captions are sufficiently clean and informative. Finally, we contextualize cross-modal alignment, introducing unimodal and multimodal baselines on various datasets. Our results show that the cross-modal representational alignment of unimodal models can reach and even surpass that of explicitly aligned vision-language models, but remains substantially behind the alignment observed for models of the same modality.
Show more
View full details

Spectral Re-Basin for Linear Mode Connectivity

Ya-Wei Eileen Lin ⋅ Thomas Dagès ⋅ Daniel Herbst ⋅ Daniel Cremers ⋅ Stefanie Jegelka
Palais des congrès Room 241
Linear mode connectivity is often analysed through exact parameter symmetries, especially hidden-unit permutations that leave the network function unchanged. Yet, many relationships between trained networks are not captured by bijective coordinate relabelings: widths may differ, neurons may split or clone, and explicit parameter symmetries may be broken. In this paper, we introduce *spectral re-basin*, taking an operator perspective on hidden-unit geometry and studying mode connectivity through the functional geometry of hidden units. Specifically, we represent each hidden layer as a graph of *neurons* built from layerwise descriptors (weights or activations), associate it with a graph shift operator, and study two coupled objects: layerwise graph functional symmetries, which preserve the relational geometry, and intertwining correspondences across layers from two models, which align their graph structures. We show that permutations arise as a special case of this viewpoint and derive results for non-bijective constructions. Our analysis connects correspondence quality to preserved spectral modes, functional exchangeability of neurons, and bounds on endpoint distortion and linear path barriers. These findings suggest that linear mode connectivity can be governed beyond explicit parameter symmetry by broader spectral operator compatibility in hidden-unit geometry.
Show more
View full details

Constrained sample efficient reinforcement learning for heavy ion accumulation in the Low Energy Ion Ring

Borja Rodriguez Mateos ⋅ Verena Kain ⋅ Michael Schenk
The Low Energy Ion Ring (LEIR) at CERN turns long lead-ion pulses from Linac3 into the dense bunches required by the Large Hadron Collider's (LHC) high intensity ion program. It accumulates successive multi-turn injections through six-dimensional phase space painting under electron cooling. The accumulated intensity depends sensitively on nine parameters, and the optimum drifts on every relevant timescale: continuously, as the stripper foil that produces the injected charge state ages, and discontinuously, as the machine is recommissioned every yearly campaign or after critical hardware faults. In this paper, we describe an operational Reinforcement Learning (RL) pipeline deployed on LEIR that treats this drift as the central design constraint and outperforms routine operational performance by more than 10\%. Using four years of archived operational data and system identification between 2023 and 2026, we first quantify the drift: generative topographic maps of the nine-dimensional setting space show that both the sampled region and the high-efficiency region move from year to year, and surrogate models fitted on one campaign transfer to the next with near zero rank correlation. We then show how to choose a low-dimensional observation robust to drift: a joint denoising autoencoder with physics-informed auxiliary heads, benchmarked against variational, quantized, and joint embedding alternatives under measured acquisition noise, compresses $\sim$136\,000 bin longitudinal Schottky spectrograms into nine latents that remain informative across campaigns. Policies are trained on per-year heteroscedastic ensemble digital twins and deployed by warm finetuning on the machine: relocking to a new campaign takes a few hundred machine cycles (1 to 2 dedicated hours of training), $\approx$33$\times$ fewer than training from scratch. Finally, exploration is framed as a constrained Markov Decision Process (MDP) with a learned, per-campaign vacuum risk cost: with a fixed Lagrange multiplier the constrained agent explores an order of magnitude more safely than its unconstrained ancestor.
Show more
View full details

On-the-fly Weight Generation: A Hypernetwork Proof of Concept on ARC-1D

Fabio J Fehr ⋅ Philip Torr
General-purpose models can adapt to many tasks from context, while specialised models can execute individual functions with substantially less capacity. Yet obtaining such specialists requires task-specific training or adaptation. We ask whether they can instead be generated directly from a few demonstrations. Using ARC-1D as a controlled testbed, we show that individual transformations admit tiny specialist models, and that a hypernetwork can generate such models directly from demonstrations. The generated parameters form a structured weight space, while the resulting specialists generalise across instances and show partial compositional generalisation to unseen combinations of known transformations, particularly without explicit task identifiers. Together, these results provide a proof of concept that few-shot task context can be compiled on-the-fly into compact executable model parameters, and that the resulting weight space can support reuse and partial compositional generalisation.
Show more
View full details

On Emergent Capabilities and Model Merging

Luca Zhou ⋅ Emanuele Rodolà
Fine-tuned checkpoints and adapters now fill public repositories, and the most common operation applied to these artifacts is model merging: arithmetic on their weights that assembles capabilities cheaply. We ask what this operation does to emergent capabilities: behaviors an artifact carries that were never an explicit training target. Studying two independent testbeds (activation oracles and emergent-misaligned models) across three model families, we find that the answer is threefold. First, merging preserves an emergent capability that both parents carry: merging two misaligned checkpoints retains most of their broad misalignment across the whole mixing range. Second, merging cannot create an emergent capability that is superadditive in its parents: no weighted merge of two single-task oracles reaches the jointly-trained oracle's auditing ability. Third, when only one parent carries the capability, merging dilutes it faster than the trained capability that accompanies it: the gap is significant in most settings. In short, emergent behaviors of an artifact do not compose the way its trained capability does.
Show more
View full details

WSI-Lab: A Pathologist-Inspired, High-Throughput RL Environment for Sequential Whole-Slide Analysis

Ahmed Alagha ⋅ Hadi Otrok ⋅ Rabeb Mizouni ⋅ Jamal Bentahar ⋅ Mahdi S. Hosseini
Whole-slide images (WSIs) are gigapixel digital pathology slides where sparse diagnostic evidence makes exhaustive high-magnification inspection computationally expensive. Pathologists instead examine selected regions across magnifications, retain global context and examination history, update their assessment as evidence accumulates, and decide where to look next and when to stop. This naturally frames WSI examination as a partially observable sequential decision problem, for which reinforcement learning (RL) provides a principled framework to learn observation, navigation, and stopping policies. Existing RL and adaptive-sampling methods largely provide task-specific solutions, while existing environments expose limited examination states that do not reflect the richer information available to a pathologist during sequential slide examination and can be costly to execute at RL scale. We introduce WSI-Lab, a reusable, Gymnasium-compatible, pathologist-inspired environment combining the current viewport, global slide context, location and examination history, and an evolving machine diagnostic-belief estimate from acquired evidence. Agents navigate across magnifications and terminate with an explicit diagnosis. Tissue-aware preprocessing, precomputed navigation structures, caching, and efficient state updates reduce repeated online work. RL ablations show that WSI-Lab's complete examination state, tissue-aware navigation, and explicit diagnosis actions support substantially stronger training than reduced variants. Under matched environment workloads, WSI-Lab is additionally up to 2.24x faster while exposing a logical observation payload up to 10.7x larger than existing environments. Code is available at https://anonymous.4open.science/r/wsi-lab-0EEC/.
Show more
View full details

Beyond Reduced-Order Control: Physics-Guided Nine-DOF Rollouts for Wave-Energy RL

Reuben Stanley
Wave energy is a substantial renewable resource, but commercial devices must extract energy competitively under changing waves. For the coupled D-Spar con- verter this creates two linked problems: predicting nonlinear device motion ac- curately and choosing power-take-off (PTO) damping quickly enough for con- trol. High-fidelity numerical models retain coupled motion, radiation memory and nonlinear forces, but are expensive to evaluate repeatedly. Rather than further reducing the device dynamics, we learn the expensive nonlinear-force correction inside the full nine-degree-of-freedom model. The remaining excitation, radia- tion, mechanics, PTO and RK4 calculations stay explicit. On development data, a free-recursive 10-s forecast obtains 0.382% macro NRMSE over all displace- ment and velocity channels. A rank-5 proper orthogonal decomposition (POD5) decoder then reconstructs a 15-action damping–energy curve from only three roll- outs, with 0.202% RMS-relative curve error, 100/110 exact sampled optima, exact- or-adjacent selection in every case and 0.0596% maximum regret. The complete measured K3+POD5 controller is not yet 10-Hz real time. Residual RL benefits from POD context but remains 0.606% below greedy. By contrast, exact two-wave sequence MPC gains 4.21% over matched wave-synchronous greedy and 3.57% over canonical 10-Hz greedy across three development seeds, while losing on one seed. These perfect-preview results motivate learning a compact wave-level policy from the MPC value landscape; they do not yet establish RL superiority, unseen- seed performance or real-time deployment.
Show more
View full details

Batch-to-Streaming Deep Reinforcement Learning for Continuous Control

Riccardo De Monte ⋅ Matteo Cederle ⋅ Gian Antonio Susto
State-of-the-art deep reinforcement learning (RL) methods have achieved remarkable performance in continuous control tasks, yet their computational complexity is often incompatible with the constraints of resource-limited hardware, due to their reliance on replay buffers, batch updates, and target networks. The emerging paradigm of streaming deep RL addresses this limitation through purely online updates. Existing streaming methods, however, are studied in isolation, trained from scratch until convergence, whereas on-device learning is most useful as a continuation of a pre-trained policy. In this work, we propose two novel streaming deep RL algorithms, Streaming Soft Actor-Critic (S2AC) and Streaming Deterministic Actor-Critic (SDAC), designed as streaming counterparts of SAC and TD3, with performance comparable to state-of-the-art streaming baselines on standard benchmarks. Building on them, we study the batch-to-streaming transition as a problem in its own right: we show that a naive transition can cause a severe and lasting loss of the pre-trained policy, trace the failure to a mismatch between the optimizers used on either side of the switch and show that aligning the optimizers enables streaming deep RL in settings such as Sim2Real.
Show more
View full details

Where Does Alignment Emerge and What Does It Mean? A Closer Look at the Platonic Representation Hypothesis

Julia Herbst ⋅ Simon Roschmann ⋅ Quentin Bouniot ⋅ Zeynep Akata
The Platonic Representation Hypothesis (PRH) suggests that neural networks are converging toward similar representation spaces. As language models become more capable, their representations are aligning with those of vision models. However, recent work has questioned whether this trend continues to hold for newer language models. Moreover, it has been shown that representation similarity metrics can be confounded by model depth and width, requiring appropriate baselines for interpretation. In this work, we first investigate where alignment emerges between unimodal models across model layers and representation dimensions. Next, we find that the PRH capability–alignment trend persists for newer models as long as the captions are sufficiently clean and informative. Finally, we contextualize cross-modal alignment, introducing unimodal and multimodal baselines on various datasets. Our results show that the cross-modal representational alignment of unimodal models can reach and even surpass that of explicitly aligned vision-language models, but remains substantially behind the alignment observed for models of the same modality.
Show more
View full details

In-Context Pure Exploration in Continuous Decision Spaces

Alessio Russo ⋅ Yin-Ching Lee ⋅ Ryan Welch ⋅ Aldo Pacchiano
In active sequential testing, also termed pure exploration, a learner is tasked with the goal to adaptively acquire information so as to identify an unknown ground-truth hypothesis with as few queries as possible. This problem has several motivating applications, including Best-Arm Identification (BAI) in bandits, where actions index hypotheses, and generalized search problems, where strategically chosen queries reveal partial information about a hidden label. In many modern settings, however, the hypothesis, or recommendation space, is continuous: for example, identifying a near optimal action in a continuous-armed bandit, localizing an $\epsilon$-ball contained in a target region, or estimating the minimizer of a function from noisy observations. Existing methods are predominantly frequentist and model-specific, while learned approaches have been limited to finite recommendation spaces. We introduce C-ICPE, a theory-guided learned model for Bayesian fixed-confidence pure exploration with continuous recommendations. C-ICPE meta-trains sequential architectures over a task prior to jointly learn exploration, stopping and recommendations strategies. At inference time, it actively gathers evidence on tasks and identifies an -optimal recommendation without parameter updates.
Show more
View full details

Generative Language Modeling for Antibody CDR Grafting and Alignment

Ferran Gonzalez ⋅ Oliver M. Turnbull ⋅ Maliha Sultana ⋅ Lorena Roldán Martin ⋅ Rohan J Kumar ⋅ Tom Diethe ⋅ Rebecca Croasdale-Wood ⋅ Charlotte Deane ⋅ Dino Oglic
Antibodies recognise their targets through hypervariable complementarity-determining regions (CDRs), which are interleaved with conserved frameworks in sequence space, making *de novo* CDR design an infilling problem. Autoregressive models generate residues left-to-right, which precludes full framework context during CDR generation and conflates framework and CDR likelihoods, leaving no natural prompt–response interface for feedback to steer generation. We present GenCDR, a family of LLaMa-based autoregressive language models that read all frameworks as a conditioning prompt and generate all CDRs jointly as a variable-length response, making CDR likelihoods a clean, separable target for reward attribution. The family comprises *IgGenCDR*, *p-IgGenCDR*, and *NanoGenCDR*, trained on unpaired, paired, and nanobody chains, respectively. GenCDR achieves the highest CDR recovery among autoregressive models and produces natural, diverse, human-like CDRs whose likelihoods correlate with fitness and developability assays. The prompt–response boundary also enables principled alignment: reward signals for binding affinity, expression, or developability can be composed to steer CDR generation. Over four rounds of alignment against antibody–antigen co-folding and developability objectives, we find that NanoGenCDR can reach *in silico* structural interface metrics competitive with those of a structure-conditioned diffusion pipeline at roughly half the sampling budget, with more natural, developable designs. The same interface can be extended to integrate experimental feedback, opening a path to closed-loop antibody *de novo* design.
Show more
View full details

Beyond Feature-Wise Interpretability: Discovering Causal Conjunctions in LLMs

Teodora Kamova ⋅ Jilles Vreeken ⋅ Nils Philipp Walter ⋅ Joscha Cüppers
Sparse autoencoders (SAEs) provide interpretable feature dictionaries for language-model activations, yet model behaviors are rarely explained by a single feature in isolation. We investigate whether conjunctions of SAE features, which we call patterns, can serve as causal mechanisms of LLM behavior. We introduce RegresNaps, a differentiable pattern miner that discovers conjunctions of binary features predictive of a continuous target. A binarized autoencoder learns candidate conjunctions, a mixture-density head relates their occurrence to the target, and held-out statistical tests filter and calibrate the resulting patterns. We apply RegresNaps to Gemma-3-27B-it with GemmaScope-2 SAEs across four behaviors: factual recall, distraction, instructed deception, and refusal. We find that the causal role of a pattern depends on the target behavior. Patterns associated with factual recall mostly carry answer information, so ablating them more often disrupts a correct answer than fixes a wrong one, and only a small subset has a corrective effect. By contrast, distraction patterns support more direct mechanical control: intervening on selected patterns repairs derailments with up to a 35.5% net gain. Interventions on deception patterns restore 22.3-55.1% of deceptive answers. Refusal patterns exhibit the strongest bidirectional control, with ablation bypassing up to 99% of refusals as measured by an independent safety classifier, while activation can also induce refusal on benign prompts. Feature-level audits show that these conjunctions combine semantic state, task context, and output-level features. Differentiable pattern mining therefore provides a scalable route from sparse feature dictionaries to behavior-specific mechanistic hypotheses, while causal intervention determines whether a mined pattern causes or merely accompanies a behavior.
Show more
View full details

Beyond Feature-Wise Interpretability: Discovering Causal Conjunctions in LLMs

Teodora Kamova ⋅ Jilles Vreeken ⋅ Nils Philipp Walter ⋅ Joscha Cüppers
Palais des congrès Room 241
Sparse autoencoders (SAEs) provide interpretable feature dictionaries for language-model activations, yet model behaviors are rarely explained by a single feature in isolation. We investigate whether conjunctions of SAE features, which we call patterns, can serve as causal mechanisms of LLM behavior. We introduce RegresNaps, a differentiable pattern miner that discovers conjunctions of binary features predictive of a continuous target. A binarized autoencoder learns candidate conjunctions, a mixture-density head relates their occurrence to the target, and held-out statistical tests filter and calibrate the resulting patterns. We apply RegresNaps to Gemma-3-27B-it with GemmaScope-2 SAEs across four behaviors: factual recall, distraction, instructed deception, and refusal. We find that the causal role of a pattern depends on the target behavior. Patterns associated with factual recall mostly carry answer information, so ablating them more often disrupts a correct answer than fixes a wrong one, and only a small subset has a corrective effect. By contrast, distraction patterns support more direct mechanical control: intervening on selected patterns repairs derailments with up to a 35.5% net gain. Interventions on deception patterns restore 22.3-55.1% of deceptive answers. Refusal patterns exhibit the strongest bidirectional control, with ablation bypassing up to 99% of refusals as measured by an independent safety classifier, while activation can also induce refusal on benign prompts. Feature-level audits show that these conjunctions combine semantic state, task context, and output-level features. Differentiable pattern mining therefore provides a scalable route from sparse feature dictionaries to behavior-specific mechanistic hypotheses, while causal intervention determines whether a mined pattern causes or merely accompanies a behavior.
Show more
View full details

MARL-based System Optimization for Wildfire Sensing Networks

River Bartz ⋅ Ramakrishnan Durairajan ⋅ Maggie Wigness
Terrestrial-based systems for early wildfire detection often use cameras deployed in remote environments, where images from these cameras are routed through the network and processed by ML models. However, determining an optimal policy for how to route and process collected images is challenging due to highly variable energy availability and network capacity at the edge. Existing methods typically adopt static approaches for determining such policies. They also rely on centralized execution, introducing a potential single point of failure. Multi-agent reinforcement learning (MARL) provides an attractive alternative. However, learned policy performance for MARL-based solutions is dependent on the quality of reward modeling, environment diversity, re-start strategies, and reward credit assignment, all of which can be difficult to define and if done poorly can lead to inconsistent and sub-optimal performance in live deployments. To address these challenges, we propose MARL-based System Optimization for Wildfire Sensing Networks (MARL-OptWSN), a framework for optimizing highly variable image-based wildfire sensing networks. MARL-OptWSN uses a custom training regime built around a Gurobi-based oracle, while leveraging decentralized execution to improve resilience and adapt to changing environmental conditions. Our training approach combines progressive training, oracle guidance, and an agent-specific blame system that identifies decisions responsible for poor outcomes. Empirical experiments show that our approach achieves near-oracle performance with substantially lower decision time, while retaining resilience under highly variable and constrained conditions that negatively affect centralized approaches.
Show more
View full details

Weight Oracles: Reading Neural Network Weights with Language Models

Krishna Kabra ⋅ Constantin Venhoff ⋅ Christian Schroeder de Witt
Interpretability methods for neural networks are predominantly reactive: they analyse activations produced during specific forward passes, requiring known inputs to find hidden capabilities such as backdoors. We propose Weight Oracles, fine-tuned language models that diagnose properties of a target network by reading its raw weights directly, without behavioural testing. We investigate this paradigm in two phases. Phase I establishes feasibility: through a staged curriculum and an external chain-of-computation that delegates composition to deterministic code, an explainer LLM learns the weight-reading primitive underlying the forward pass of small transformers, simulating them end-to-end, achieving 99% holdout accuracy on unseen targets. Phase II repurposes this infrastructure for safety auditing. We train an oracle on natural language diagnostic questions about weight anomalies using only benign pathologies as training signal, and evaluate it zero-shot on backdoors absent from training. The oracle achieves AUROC 0.93 on attention routed backdoors and 0.81 across a diversified threat distribution including stealth and adversarially regularized variants. Hand-crafted statistical detectors are sharp on the threat models they implicitly target but collapse on threat-model shift, while the oracle remains uniformly competent across attack types. Scaling to realistic model sizes remains the principal open challenge.
Show more
View full details

MA-BED: Coordinating Multiple Actors in Bayesian Experimental Design

Marcel Hedman
To date, Bayesian experimental design (BED) has largely studied how a single adaptive policy should choose experiments. Yet many real-world data-acquisition systems distribute experimental capacity across multiple actors, such as laboratories, field teams, or decentralised robots. We introduce multi-actor Bayesian experimental design (MA-BED) as a general paradigm for BED in which experimental decisions are distributed across multiple actors, and within this paradigm focus on a fully cooperative setting with decentralised execution. In this regime, actors share an inferential objective but choose experiments using only their local histories, creating a coordination problem: individually informative policies may collectively acquire redundant information. We formalise this failure through an information-theoretic coordination gap and show that, under conditionally independent acquisition, it is exactly characterised by the total correlation among actor histories. To address this problem, we adapt centralised training with decentralised execution (CTDE) to BED, jointly optimising locally executable policies against the information contained in their pooled experimental histories. We instantiate this framework with DAD-based algorithms trained using a contrastive information objective. Empirically, joint training reduces coordination failure, induces greater actor specialisation under heterogeneous experimental costs, and outperforms independently trained policies without requiring communication at execution. These results establish coordination as an important challenge in Bayesian experimental design when data acquisition is distributed across multiple actors.
Show more
View full details

Many Biases, One Circuit: Post-Training Tunes a Pretrained Cue-Routing Mechanism

Prakhar Gupta ⋅ Terry J. C. Zhang ⋅ Florent Draye ⋅ Bernhard Schölkopf ⋅ Zhijing Jin
Palais des congrès Room 241
Language models often abandon a correct answer when a prompt cue points at a wrong one. This failure appears in sycophancy and several related cue-induced biases. We ask where this behavior comes from during post-training. Across six model and recipe lineages, these biases converge on a recurring reader-writer attention circuit that is already causally active before alignment. Generic assistant adaptation makes chat-formatted prompts engage this circuit, even without any sycophancy-targeted data. Later post-training stages then change how strongly the circuit influences the answer. Across public Tulu-3 and OLMo-2 checkpoints, the circuit's routing tracks each stage's change in cue-following, while causally raising the model's confidence does not reduce cue-following. Most importantly, swapping the naturally learned routing patterns of one training stage into another moves the behavior toward the donor stage while preserving uncued answers. The same circuit also contributes causally to two-turn retraction. Our results suggest that post-training does not build a new sycophancy mechanism. It changes when pretrained cue-routing machinery is used and how strongly it affects the answer.
Show more
View full details

Many Biases, One Circuit: Post-Training Tunes a Pretrained Cue-Routing Mechanism

Prakhar Gupta ⋅ Terry J. C. Zhang ⋅ Florent Draye ⋅ Bernhard Schölkopf ⋅ Zhijing Jin
Language models often abandon a correct answer when a prompt cue points at a wrong one. This failure appears in sycophancy and several related cue-induced biases. We ask where this behavior comes from during post-training. Across six model and recipe lineages, these biases converge on a recurring reader-writer attention circuit that is already causally active before alignment. Generic assistant adaptation makes chat-formatted prompts engage this circuit, even without any sycophancy-targeted data. Later post-training stages then change how strongly the circuit influences the answer. Across public Tulu-3 and OLMo-2 checkpoints, the circuit's routing tracks each stage's change in cue-following, while causally raising the model's confidence does not reduce cue-following. Most importantly, swapping the naturally learned routing patterns of one training stage into another moves the behavior toward the donor stage while preserving uncued answers. The same circuit also contributes causally to two-turn retraction. Our results suggest that post-training does not build a new sycophancy mechanism. It changes when pretrained cue-routing machinery is used and how strongly it affects the answer.
Show more
View full details

Dynamic Plasma Shape Control Robust to Diagnostic Dropout

Dmitrii Sorokin ⋅ Maksim Stokolesov ⋅ Alexander A Granovskiy ⋅ Igor Prokofyev ⋅ Evgeny Adishchev ⋅ Maxim Nurgaliev ⋅ Eduard Khairutdinov ⋅ Georgy Subbotin ⋅ Randall Clark ⋅ Dmitri M Orlov
Plasma shape control in tokamaks requires a real-time controller that tracks dynamically changing shape targets while tolerating diagnostic failures. Classical approaches decompose the problem into equilibrium reconstruction followed by a linear controller, and assume a fixed, fully operational sensor set. We present a reinforcement learning agent that addresses both limitations simultaneously. The agent is trained in NSFsim, a high-fidelity tokamak simulator configured for DIII-D, on a curated dataset of 120 experimental plasma shapes. The shape targets are resampled as random step changes every 0.25\,s, exposing the agent to diverse transitions across the full shape envelope. At test time the agent zero-shot tracks dynamic shape sequences; on a held-out static configuration in simulation it achieves a mean shape error of 2.01\,cm, and dynamic trajectory following is demonstrated qualitatively in simulation and on the physical device. Diagnostic dropout randomly masks 30\% of magnetic sensors per episode, yielding a single policy robust to a distribution of sensor subsets used for tokamak operation, without backup controllers or mode-switching logic. An asymmetric actor-critic architecture with privileged equilibrium information improves value estimation under partial observability; an auxiliary shape reconstruction head on the actor enables end-to-end shape reconstruction from raw diagnostics and serves as an interpretability tool for policy analysis. The policy transfers to experimental DIII-D shots, where it directly commands the coil actuators on two dynamic shape maneuvers, and to the independent GSevolve simulator.
Show more
View full details

Autonomous Focusing of a Capillary Optic for Angle-Resolved Photoemission Spectroscopy

Hippolyte Bourel ⋅ Alfred J Jones ⋅ Søren Ulstrup ⋅ Jill Miwa ⋅ Philip Hofmann ⋅ Panagiotis Karras
Low-dimensional optimization problems in complex experimental environments are common and often take the form of black-box optimization problems. We present a protocol for automated alignment of a capillary optic that focuses extreme-ultraviolet light for angle-resolved photoemission spectroscopy (ARPES). This alignment is a demanding procedure typically performed by expert beamline scientists and is generally inaccessible to non-expert users. In spatially resolved ARPES, a small beam spot is essential for measurements of challenging materials and functional devices. We compare several optimization methods in simulations based on experimental data and demonstrate successful automated focusing in experiments at the SGM4 microARPES beamline of the ASTRID2 synchrotron radiation source in Aarhus.
Show more
View full details