论文检索

Theory · Reinforcement Learning and Planning

Amogh Palasamudram, Jakub Svoboda, Krishnendu Chatterjee, Suguman Bansal

{\em Reinforcement learning} (RL) for {\em reachability specifications} is fundamental in sequential decision-making, yet theoretical guarantees remain less explored. A recent work achieves {\em asymptotic convergence} to optimal policies. However, this approach provides limited insight into convergence dynamics. In this work, we present an alternative approach that provides deeper theoretical insights into convergence. Our approach builds on {\em PAC learning} with assumptions. PAC learning guarantees near-optimal policies with high confidence in finite time but requires knowing internal MDP parameters like minimum transition probability. We argue that while these parameters are unknown in RL, they can be iteratively refined and estimated with increasing accuracy. By iteratively satisfying PAC conditions, we show that exact optimality can be achieved in the limit. Empirical evaluations on standard benchmarks validate our theoretical insights into convergence dynamics.

Deep Learning · Theory

Shwan Ashrafi, Dan Roth

Retrieval-augmented generative agents rely on retrieval for grounding, yet are typically evaluated on a query-by-query basis. This isolates interactions that are geometrically coupled in a shared embedding space. For example, we show that the high document density required to serve majority interests (e.g., generic "Crime" movies) can geometrically overcrowd the retrieval neighborhood of a semantically similar minority (e.g., "Film Noir"), effectively expelling minority content from top-$k$ results. We introduce a formal framework to analyze how such *goal collisions* in dense retrieval induce fundamental performance limits and emergent fairness issues inherent to spatial crowding. In our static analysis, we demonstrate that for a fixed embedding space, a phase transition occurs where minority user goals suffer a catastrophic collapse in performance as the density of majority goals increases. We then extend this to a dynamic model and derive a non-linear Fokker-Planck equation that governs the evolution of document embeddings as the agent updates them to maximize retrieval accuracy. Our analysis reveals that this local relevance objective triggers an emergent global mechanism that systematically marginalizes minority interests. We prove that such objectives drive the system to self-organize into a state that exclusively serves majority interests. These results provide a theoretical foundation for understanding a critical grounding failure mode in retrieval-augmented agents.

Deep Learning · Generative Models and Autoencoders

Davide Gallon, Philippe von Wurstemberger, Patrick Cheridito, Arnulf Jentzen

We propose a methodology that combines generative latent diffusion models with physics-informed machine learning to generate solutions of parametric partial differential equations (PDEs) conditioned on partial observations, which includes, in particular, forward and inverse PDE problems. We learn the joint distribution of PDE parameters and solutions via a diffusion process in a latent space of scaled spectral representations, where Gaussian noise corresponds to functions with controlled regularity. This spectral formulation enables significant dimensionality reduction compared to grid-based diffusion models and ensures that the induced process in function space remains within a class of functions for which the PDE operators are well defined. Building on diffusion posterior sampling, we enforce physics-informed constraints and measurement conditions during inference, applying Adam-based updates at each diffusion step. We evaluate the proposed approach on Poisson, Helmholtz, and incompressible Navier--Stokes equations, demonstrating improved accuracy and computational efficiency compared with existing diffusion-based PDE solvers, which are state of the art for sparse observations. Code is available at https://anonymous.4open.science/r/PISD-32A0/.

Applications · Computer Vision

David Acuna, Chao-Han Huck Yang, Yuntian Deng, Jaehun Jung, Ximing Lu, Prithviraj Ammanabrolu, Hyunwoo Kim, Yuan-Hong Liao, Yejin Choi

Despite rapid progress, multimodal reasoning still lacks a systematic approach to synthesize large-scale vision-centric datasets beyond visual math. We introduce a framework able to synthesize vision-centric problems spanning diverse levels of complexity, and the resulting dataset with over 1M high-quality problems including: reasoning traces, preference data, and instruction prompts supporting SFT, offline and online RL. Our vision-centric synthesis framework uses a two-stage process focusing on: (1) generating diverse verifiable questions from existing images at scale, and (2) creating complex compositional visual problems by merging simpler questions. Remarkably, finetuning Qwen2.5-VL-7B on our data outperforms existing open-data baselines across evaluated vision-centric benchmarks, and our best configurations match or surpass strong closed-data models such as MiMo-VL-7B-RL on V*Bench, CV-Bench and MMStar-V. Notably, despite being entirely vision-centric, our data transfers positively to text-only reasoning (MMLU-Pro, +3.7%) and audio reasoning (MMAU, +1.32%), demonstrating its effectiveness. Similarly, despite containing no embodied visual data, we observe notable gains (NiEH, +8.8%) when evaluating open-ended embodied QA. Lastly, we use our data to comprehensively analyze at scale (1M+) the entire VLM post-training pipeline showing that (i) SFT on high-quality data with cognitive behaviours on reasoning traces is essential to scale online RL, (ii) offline RL could match online RL’s performance while disaggregating compute demands, and, (iii) SFT on high quality data also improve out-of-domain, cross-modality transfer.

Applications · Language, Speech and Dialog

Jiashuo Sun, Pengcheng Jiang, Saizhuo Wang, Jiajun Fan, Heng Wang, Siru Ouyang, Ming Zhong, Yizhu Jiao, Chengsong Huang, Xueqiang Xu 等

Retrieval-Augmented Generation (RAG) systems remain brittle under realistic retrieval noise, even when the required evidence appears in the top-$K$ results. A key reason is that retrievers and rerankers optimize solely for relevance, often selecting either trivial, answer-revealing passages or evidence that lacks the critical information required to answer the question, without considering whether the evidence is suitable for the generator. We propose \texttt{BAR-RAG}, which reframes the reranker as a boundary-aware evidence selector that targets the generator’s Goldilocks Zone—evidence that is neither trivially easy nor fundamentally unanswerable for the generator, but is challenging yet sufficient for inference and thus provides the strongest learning signal. \texttt{BAR-RAG} trains the selector with reinforcement learning using generator feedback, and adopts a two-stage pipeline that fine-tunes the generator under the induced evidence distribution to mitigate the distribution mismatch between training and inference. Experiments on knowledge-intensive question answering benchmarks show that \texttt{BAR-RAG} consistently improves end-to-end performance under noisy retrieval, achieving an average gain of 10.3\% over strong RAG and reranking baselines while substantially improving robustness.

Deep Learning · Theory

Francesco Cagnetta, Allan Raventos, Surya Ganguli, Matthieu Wyart

Despite the fact that experimental neural scaling laws have substantially guided empirical progress in large-scale machine learning, no existing theory can quantitatively predict the exponents of these important laws for any modern LLM trained on any natural language dataset. We provide the first such theory in the case of data-limited scaling laws. We isolate two key statistical properties of language that {\it alone} can predict neural scaling exponents: (i) the decay of pairwise token correlations with time separation between token pairs, and (ii) the decay of the next-token conditional entropy with the length of the conditioning context. We further derive a simple formula in terms of these statistics that predicts data-limited neural scaling exponents from first principles {\it without any} free parameters or synthetic data models. Our theory exhibits a remarkable match with experimentally measured neural scaling laws obtained from training GPT-2 and LLaMA style models from scratch on two qualitatively different benchmarks, TinyStories and WikiText.

Deep Learning · Theory

Hee-Sung Kim, Sungyoon Lee

Recent theoretical analyses of multi-pathway Deep Linear Networks, typically grounded in Gradient Flow, predict a "winner-takes-all" specialization in which path symmetry breaks and each feature concentrates in a single pathway. In this work, we show that discrete Gradient Descent with a large step size reproduces the initial, depth-driven symmetry breaking but ultimately overrides this tendency due to its catapults at the Edge of Stability. In this regime, GD exhibits an implicit preference for low-curvature minima. Since we prove that splitting singular values across pathways minimizes sharpness, large-step GD—driven by its implicit preference for such flat minima—forces a subsequent re-balancing phase: iterates escape sharp, sparse configurations for stable, balanced solutions. Together, these results clarify how architectural depth shapes pathway competition while explaining why GD with large step size ultimately favors shared representations rather than permanent pathway monopolization.

Deep Learning · Large Language Models

Tao Feng, Tianyang Luo, Jingjun Xu, Zhigang Hua, Yan Xie, Shuang Yang, Ge Liu, Jiaxuan You

Experience learning has achieved promising results in enhancing LLM agent planning and reasoning by integrating past interactions as reusable knowledge. However, existing methods remain confined to explicit text space---retrieving experiences via semantic similarity and concatenating them into the context window, leading to substantial token overhead and a decoupled architecture that separates retrieval from generation. To address these limitations, we propose \method, a framework that enables LLM agents to learn from experience via latent retrieval-augmented generation, without requiring a separate RAG module. \method encodes experiences using the LLM's own hidden states, retrieves relevant experiences directly in latent space at each decoding step, and integrates them through cross-attention aggregation and gated residual mechanisms. The entire pipeline is optimized end-to-end with reinforcement learning, supporting both generative and ranking tasks. We evaluate \method on 13 diverse tasks spanning question answering, reasoning, coding, scientific prediction, and recommendation. Results demonstrate that: (1) \method achieves state-of-the-art on 12 out of 13 tasks, outperforming the strongest baseline by over 6.8\%; (2) \method maintains token efficiency comparable to non-retrieval baselines while text-based retrieval methods require 1.5--2$\times$ more tokens; and (3) \method exhibits superior cross-domain generalization, outperforming the strongest baseline by 16.32\% under zero-shot transfer and 15.21\% under few-shot transfer.

Social Aspects · Alignment

Rakshit Trivedi, Natasha Jaques, Logan Cross, Alexander Vezhnevets, Joel Z Leibo

AI's central challenge is shifting from capability to coexistence. The dominant paradigm in AI research focuses on developing powerful agents under stationary-environment assumptions, treating the world as an exogenous source of feedback. This position paper argues that a solipsistic superintelligence---an extremely capable solver of stationary problems---is unlikely to be cooperative. Deployment induces endogenous nonstationarity: other agents adapt, producing best-response dynamics that reshape the environment the AI was trained to navigate. The result is a train--test--deploy gap where historical distributions diverge from deployment realities; the more aggressively a solipsistic superintelligence exploits historical regularities, the faster it renders them obsolete. Cooperation is therefore not an added capability but an equilibrium property that solipsistic superintelligence cannot guarantee. We call for a multi-agent-first research paradigm treating strategic interdependence as a core design principle, alongside dynamic evaluation: testbeds where distributions are generated by adaptive counterparties, and metrics prioritizing equilibrium stability over single-score task success.

General Machine Learning · Unsupervised and Semi-supervised Learning

Se Yoon Lee, Jae-kwang Kim

Obtaining high-quality labels is costly, whereas unlabeled covariates are often abundant, motivating semi-supervised inference methods with reliable uncertainty quantification. Prediction-powered inference (PPI) leverages a machine-learning predictor trained on a small labeled sample to improve efficiency, but it can lose efficiency under model misspecification and suffer from coverage distortions due to label reuse. We introduce Machine‑Learning‑Assisted Generalized Entropy Calibration (MEC), a cross‑fitted, calibration‑weighted variant of PPI. MEC improves efficiency by reweighting labeled samples to better align with the target population, using a principled calibration framework based on Bregman projections. This yields robustness to affine transformations of the predictor and relaxes requirements for validity by replacing conditions on raw prediction error with weaker projection‑error conditions. As a result, MEC attains the semiparametric efficiency bound under weaker assumptions than existing PPI variants. Across simulations and a real‑data application, MEC achieves near‑nominal coverage and tighter confidence intervals than CF‑PPI and vanilla PPI.

General Machine Learning · Evaluation

Yuheng Yang, Siqi Zhu, Tao Feng, Ge Liu, Jiaxuan You

Large Language Models (LLMs) can be seen as compressed knowledge bases, but it remains unclear what knowledge they truly contain and how far their knowledge boundaries extend. Existing benchmarks are mostly static and provide limited support for systematic knowledge probing. In this paper, we propose an interactive agentic framework to systematically extract and quantify the knowledge of LLMs. Our method includes four adaptive exploration policies to probe knowledge at different granularities. To ensure the quality of extracted knowledge, we introduce a three-stage knowledge processing pipeline that combines vector-based filtering to remove exact duplicates, LLM-based adjudication to resolve ambiguous semantic overlaps, and domain-relevance auditing to retain valid knowledge units. Through extensive experiments, we find that recursive taxonomy is the most effective exploration strategy. We also observe a clear knowledge scaling law, where larger models consistently extract more knowledge. In addition, we identify a Pass@1-versus-Pass@k trade-off: domain-specialized models achieve higher initial accuracy but degrade rapidly, while general-purpose models maintain stable performance during extended extraction. Finally, our results show that differences in training data composition lead to distinct and measurable knowledge profiles across model families.

Deep Learning · Generative Models and Autoencoders

Isma Hadji, Enrique Sanchez, Adrian Bulat, Brais Martinez, Georgios Tzimiropoulos

We introduce a multi-scale Image Super Resolution (ISR) method building on recent advances in Visual Auto-Regressive (VAR) modeling. Recently, VAR models challenged the dominance of diffusion-based models by adopting a next-scale prediction paradigm. Specifically, VAR models iteratively estimate the residual in latent space between gradually increasing image scales, an approach that aligns perfectly with our target ISR task. Previous work taking advantage of this synergy, suffer from two main shortcomings. First, due to the limitations of the residual quantizers used in VAR models, they typically only generate images at a predefined fixed scale, i.e. they fail to map intermediate outputs to the corresponding image scales. Also, to achieve better performance, they rely on large backbones and either external VLM for guidance, or a large corpus of external carefully annotated data. To address both shortcomings, we introduce two novel components to the VAR training for ISR, aiming at increasing its flexibility and reducing its complexity. In particular, we introduce a) a Hierarchical Image Tokenization (HIT) approach with a multi-scale image tokenizer that progressively represents images at different scales while enforcing token overlap across scales, and b) a Direct Preference Optimization (DPO) regularization term that, relying solely on the (LR,HR) pair, encourages the transformer to produce the latter over the former. The resulting model can denoise the LR image and super-resolve at different upscale factors in a single forward pass, and achieves state-of-the-art results using a relatively small model (300M params vs ~1B params of VARSR), and without the need of external training data.

General Machine Learning · Supervised Learning

Yongxiang Tang, Chao Wang, Jincheng Lu, Yanhua Cheng, Xialong Liu, Peng Jiang

Learning to rank can be cast as a probabilistic modeling problem over permutations, where the goal is to estimate the likelihood of an observed total ordering of items. This formulation naturally involves full-order probabilities of the form $\mathbb{P}(\mathrm{z}_1 < \cdots < \mathrm{z}_n)$, whose exact computation and optimization are intractable due to the factorial growth of the permutation space with respect to list size. We introduce the *Full-Order Bound* (FOB), a tractable lower bound on the probability of an observed ordering, constructed from a subset of ordering constraints that factorizes across items while avoiding low-dimensional surrogate objectives and preserving order-reversal invariance. FOB induces a convex inner tightening problem over latent cut points, which we solve efficiently during training using a *safe-region gradient ascent* (SRGA) procedure. Experiments on synthetic ranking tasks and large-scale learning-to-rank benchmarks show that FOB consistently improves performance over pairwise and listwise surrogates, Plackett--Luce style sequential factorization models, and differentiable sorting baselines. Code is available at [https://anonymous.4open.science/r/FOB_2026-46C4](https://anonymous.4open.science/r/FOB_2026-46C4).

Adil Soubki, Miles Cranmer

Symbolic Regression (SR) is the task of finding a closed-form mathematical expression that optimizes some objective. Solving this task is NP-hard. However, SR software routinely discovers accurate, interpretable models without exhaustively searching function space. Motivated by this disconnect between worst-case theory and practical success, we study SR through the lens of *parameterized complexity theory*. In particular, we reanalyze tractability with respect to practically relevant parameters like expression depth, tree size, and number of primitives used. We show that SR is actually fixed-parameter tractable (FPT) under a parametrization over expression depth or tree size, formalizing an explanation for why the bounded-complexity search of popular SR algorithms succeeds. However, SR becomes W[1]-hard when parameterized by the number of variables or primitives used, identifying selection as a source of intractability. We further find lower bounds under the exponential time hypothesis, prove approximation hardness, and rule out polynomial kernels.

Deep Learning · Generative Models and Autoencoders

Ziang Gan, Qi Zhu, Libao Zhang

Diffusion models serve as generative priors for dataset distillation, yet existing pipelines rely on per-sample update rules that evolve each synthetic image independently, limiting their ability to optimize collective set-level objectives. We propose Set-Coupled Guidance (SCG), a plug-and-play auxiliary controller that shifts from per-image to group (IPC-at-once) sampling by injecting set-symmetric feedback at each diffusion step. SCG combines spectral set-point regulation, which aligns set-level statistics to real data via empirical characteristic function matching, with cooperative kernel coupling that stabilizes joint trajectories under noisy feedback. All computations operate on lightweight descriptors extracted from predicted clean latents, adding low overhead to the base method. We provide theoretical analysis including Lyapunov descent and input-to-state stability for distributional tracking. Experiments on ImageNette, ImageWoof, ImageNet-100 and ImageNet-1K show consistent accuracy gains across multiple diffusion-based baselines.

Probabilistic Methods · Everything Else

Cristiana Diaconu, Miles Cranmer, Richard E Turner, Tanya Marwah, Payel Mukhopadhyay

Dominant approaches for modelling Partial Differential Equations (PDEs) rely on deterministic predictions, yet many physical systems of interest are inherently chaotic and uncertain. While training probabilistic models from scratch is possible, it is computationally expensive and fails to leverage the significant resources already invested in high-performing deterministic backbones. In this work, we adopt a training-efficient strategy to transform pre-trained deterministic models into probabilistic ones via retrofitting with a proper scoring rule: the Continuous Ranked Probability Score (CRPS). Crucially, this approach is architecture-agnostic: it applies the same adaptation mechanism across distinct model backbones with minimal code modifications. The method proves highly effective across different scales of pre-training: for models trained on single dynamical systems, we achieve $20–54\%$ reductions in rollout CRPS and up to $30\%$ improvements in variance-normalised RMSE (VRMSE) relative to compute-matched deterministic fine-tuning. We further validate our approach on a PDE foundation model, trained on multiple systems and retrofitted on the dataset of interest, to show that our probabilistic adaptation yields an improvement of up to $40\%$ in CRPS and up to $15\%$ in VRMSE compared to deterministic fine-tuning. Validated across diverse architectures and dynamics, our results show that probabilistic PDE modelling need not require retraining from scratch, but can be unlocked from existing deterministic backbones with modest additional training cost.

Applications · Chemistry, Physics, and Earth Sciences

Yanru Qu, Cheng-Yen Hsieh, Zaixiang Zheng, Ge Liu, Quanquan Gu

We present protein autoregressive modeling (PAR), the first multi-scale autoregressive framework for protein backbone generation via coarse-to-fine next-scale prediction. Using the hierarchical nature of proteins, PAR generates structures that mimic sculpting a statue, forming a coarse topology and refining structural details over scales. To achieve this, PAR consists of three key components: (i) multi-scale downsampling operations that represent protein structures across multiple scales during training; (ii) an autoregressive transformer that encodes multi-scale information and produces conditional embeddings to guide structure generation; (iii) a flow-based backbone decoder that generates backbone atoms conditioned on these embeddings. Moreover, autoregressive models suffer from exposure bias, caused by the training and the generation procedure mismatch, and substantially degrades structure generation quality. We effectively alleviate this issue by adopting noisy context learning and scheduled sampling, enabling robust backbone generation. Notably, PAR exhibits strong zero-shot generalization, supporting flexible human-prompted conditional generation and motif scaffolding without requiring fine-tuning. On the unconditional generation benchmark, PAR effectively learns protein distributions and produces backbones of high design quality, and exhibits favorable scaling behavior. Together, these properties establish PAR as a promising framework for protein structure generation.

Applications · Chemistry, Physics, and Earth Sciences

Yanru Qu, Cheng-Yen Hsieh, Zaixiang Zheng, Ge Liu, Quanquan Gu

We present protein autoregressive modeling (PAR), the first multi-scale autoregressive framework for protein backbone generation via coarse-to-fine next-scale prediction. Using the hierarchical nature of proteins, PAR generates structures that mimic sculpting a statue, forming a coarse topology and refining structural details over scales. To achieve this, PAR consists of three key components: (i) multi-scale downsampling operations that represent protein structures across multiple scales during training; (ii) an autoregressive transformer that encodes multi-scale information and produces conditional embeddings to guide structure generation; (iii) a flow-based backbone decoder that generates backbone atoms conditioned on these embeddings. Moreover, autoregressive models suffer from exposure bias, caused by the training and the generation procedure mismatch, and substantially degrades structure generation quality. We effectively alleviate this issue by adopting noisy context learning and scheduled sampling, enabling robust backbone generation. Notably, PAR exhibits strong zero-shot generalization, supporting flexible human-prompted conditional generation and motif scaffolding without requiring fine-tuning. On the unconditional generation benchmark, PAR effectively learns protein distributions and produces backbones of high design quality, and exhibits favorable scaling behavior. Together, these properties establish PAR as a promising framework for protein structure generation.

Peter Potaptchik, Adhi Saravanan, Abbas Mammadov, Alvaro Prat, Michael Albergo, Yee-Whye Teh

Controlling generative models—whether via inference-time steering or fine-tuning—is expensive. Control relies on estimating the value function—typically necessitating costly trajectory simulations. To eliminate this bottleneck, we introduce *Meta Flow Maps (MFMs)*, stochastic extensions of consistency models and flow maps. MFMs are trained to perform \textbf{one-step posterior sampling}, generating arbitrarily many i.i.d. draws of clean data $x_1$ from any noisy state $x_t$. Crucially, these samples are differentiable in the conditioning state $x_t$, unlocking efficient estimation of the value function gradient. We leverage this capability to enable both **inference-time steering** without inner rollouts, and unbiased, off-policy **fine-tuning** to general rewards. Among our fine-tuning and steering experiments on ImageNet, we highlight that our single-particle steered-MFM sampler outperforms a Best-of-1000 baseline across multiple rewards at a fraction of the compute.

Social Aspects · Alignment

Pushkar Mishra, Charvi Rastogi, Stephen Pfohl, Alicia Parrish, Tian Teh, Roma Patel, Mark Diaz, Ding Wang, Michela Paganini, Vinodkumar Prabhakaran 等

Ensuring the safety of Generative AI requires a nuanced understanding of pluralistic viewpoints. In this paper, we introduce a novel data-driven approach for analyzing ordinal safety ratings in pluralistic settings. Specifically, we address the challenge of interpreting nuanced differences in safety feedback from a diverse population expressed via ordinal scales (e.g., a Likert scale). We define non-parametric responsiveness metrics that quantify how raters convey broader distinctions and granular variations in the severity of safety violations. Leveraging publicly available datasets of pluralistic safety feedback as our case studies, we investigate how raters from different demographic groups use an ordinal scale to express their perceptions of the severity of violations. We apply our metrics across violation types, demonstrating their utility in extracting nuanced insights that are crucial for aligning AI systems reliably in multi-cultural contexts. We show that our approach can inform rater selection and feedback interpretation by capturing nuanced viewpoints across different demographic groups, hence improving the quality of pluralistic data collection and in turn contributing to more robust AI alignment.