论文检索

Applications · Robotics

Zhiyi Li, Peilin Wu, Xiaoshen Han, Ruojin Cai, Yilun Du

Learned world models are emerging as a powerful paradigm in robotics, offering a promising path toward task generalization, long-horizon planning, and flexible decision-making. However, prevailing approaches often operate on 2D video sequences, inherently lacking the 3D geometric understanding necessary for precise spatial reasoning and physical consistency. We introduce a **Structured 4D Latent World Model**, which predicts the evolution of a scene’s 3D structure in a structured latent space conditioned on observations and textual instructions. Our representation encodes the scene holistically and can be decoded into diverse 3D formats, enabling a more complete and physically consistent scene understanding. This structured 4D latent world model serves as a planner, generating future scenes that are translated into executable actions by a goal-conditioned inverse dynamics module. Experiments demonstrate that our model generates futures with superior visual quality, physical consistency, and multi-view coherence compared to state-of-the-art video-based planners. Consequently, our full planning pipeline achieves superior performance on complex manipulation tasks, exhibits robust generalization to novel visual conditions, and proves effective on real-world robotic platforms. Our website is available at https://icml2026-18617.github.io/.

Probabilistic Methods · Monte Carlo and Sampling Methods

Paweł Parys, Sairam Vaidya, Taylor Berg-Kirkpatrick, Loris D'Antoni

Language Models (LMs) are increasingly used in applications where generated outputs must satisfy strict semantic or syntactic constraints. Existing approaches to constrained generation fall along a spectrum: greedy constrained decoding methods enforce validity during decoding but distort the LM’s distribution, while rejection sampling (RS) preserves fidelity but wastes computation by discarding invalid outputs. Both extremes are problematic in domains such as program fuzzing, where both validity and diversity of samples are essential. We present Constrained Adaptive Rejection Sampling (CARS), an approach that strictly improves the sample-efficiency of RS without distributional distortion. CARS begins with unconstrained LM sampling and adaptively rules out constraint-violating continuations by recording them in a trie and subtracting their probability mass from future draws. This adaptive pruning ensures that prefixes proven invalid are never revisited, acceptance rates improve monotonically, and the resulting samples exactly follow the constrained distribution. In experiments on a variety of domains--e.g., program fuzzing and molecular generation--CARS consistently achieves higher efficiency--measured in the number of LM forward passes per valid sample--while also producing stronger sample diversity than both Greedy Constrained Decoding (GCD) and methods that approximate the LM's distribution.

Optimization · Everything Else

Haoming Meng, Anton Sugolov, Vardan Papyan

Deep neural networks with repeated blocks, such as transformers and ResNets, often exhibit closely related representational structure across layers that emerges with training. Motivated by this observation, we introduce *Gradient Smoothing*, a general training paradigm that couples gradient updates across blocks and admits a natural interpretation as a preconditioning method. Our framework applies structured smoothing operators to layer-wise updates, such as weighted averages and exponential moving averages, with minimal computational overhead. We evaluate Gradient Smoothing across a range of architectures and training regimes, including RL post-training of LLMs on reasoning tasks, as well as diffusion and classification with Vision Transformers. Across these settings, Gradient Smoothing consistently improves generalization performance, in addition to promoting structured representation evolution across layers. These results suggest that gradient smoothing is a simple and broadly applicable technique for improving training in modern deep networks.

Renhao Zhang, Haotian Fu, Mingxi Jia, George Konidaris, Yilun Du, Bruno C. da Silva

We propose Parameterized Diffusion Policy (PDP), a framework that learns a diffusion policy parameterized in a smooth continuous space. By structuring a latent manifold such that distances between latents' values reflect the semantic similarity of physical trajectories, we transform diffusion from a mechanism of stochastic diversity into a precise tool for behavior steering. Our approach also enables smooth interpolation between known strategies and efficient generalization to novel constraints without the need to update policy weights. We demonstrate that PDP significantly improves adaptation performance on complex multimodal benchmarks in both simulation and real-robot hardware compared to regular diffusion policy, particularly in scenarios requiring the discovery of novel behaviors.

Theory · Learning Theory

Sanyam Agarwal, Pranjal Dutta, Markus Bläser

Mixtures of Bernoulli product distributions are a simple and widely used latent-variable model, with applications in e.g.\ recommendation systems, crowdsourcing, and medical data analysis. We consider the problem of reconstructing the mixture parameters from oracle access to its probability generating polynomial (PGP), for instance represented by a probabilistic generating circuit (PGC). We show that the parameters are uniquely identifiable for almost all mixtures, and give a randomized algorithm that exactly recovers the mixture weights and component marginals for mixtures of $r$ Bernoulli product distributions over $n$ variables using only $O(nr^2)$ oracle queries. The algorithm repeatedly applies restrictions to $O(r)$ variables, extracts low-degree coefficients, and then recovers the parameters using a moment-based tensor decomposition. To the best of our knowledge, this is the {\em first} exact reconstruction algorithm in this PGP oracle model with query complexity linear in $n$ and polynomial in $r$.

Deep Learning · Large Language Models

Dylan Zhang, Yufeng Xu, Haojin Wang, Qingzhi Chen, Hao Peng

Post-training of reasoning LLMs is a holistic process that typically consists of an offline SFT stage followed by an online reinforcement learning (RL) stage. However, SFT is often optimized in isolation to maximize SFT performance alone. We show that, after identical RL training, models initialized from stronger SFT checkpoints can significantly underperform those initialized from weaker ones. We propose PEAR ($\textbf{P}$olicy $\textbf{E}$valuation–inspired $\textbf{A}$lgorithm for Offline Learning Loss $\textbf{R}$eweighting), an SFT-stage method that corrects this mismatch and better prepares the model for RL. PEAR uses importance sampling to reweight the SFT loss, with three variants operating at the token, block, and sequence levels. It can be used to augment standard SFT objectives and incurs little additional training overhead once probabilities for the offline data are collected. We conduct controlled experiments on verifiable reasoning games and mathematical reasoning tasks on Qwen2.5/3 and DeepSeek-distilled models. PEAR consistently improves post-RL performance over canonical SFT, with pass@8 gains up to a 14.6% on AIME-2025. Our results suggest that PEAR is an effective step toward more holistic LLM post-training by designing and evaluating SFT with downstream RL in mind rather than in isolation.

Deep Learning · Other Representation Learning

Davide Marincione, Michele Mancusi, Giorgio Strano, Luca Cerovaz, Donato Crisostomi, Roberto Ribuoli, Emanuele Rodolà

Stem retrieval, the task of matching missing stems to a given audio submix, is a key challenge currently limited by models that discard temporal information. We introduce PHALAR, a contrastive framework achieving a relative accuracy increase of up to $\sim 70\%$ over the state-of-the-art while requiring $<$50\% of the parameters and a 7$\times$ training speedup. By utilizing a Learned Spectral Pooling layer and a complex-valued head, PHALAR enforces pitch-invariant and phase-equivariant biases. PHALAR establishes new retrieval state-of-the-art across MoisesDB, Slakh, and ChocoChorales, correlating significantly higher with human coherence judgment than semantic baselines. Finally, zero-shot beat tracking and linear chord probing confirm that PHALAR captures robust musical structures beyond the retrieval task.

Applications · Robotics

Shuxin Cao, Liquan Wang, Walker Byrnes, Yiye Chen, Yilun Du, Animesh Garg

In this paper, we identify a semantic decomposition in robot action sequences, separating task-level motion intent from execution-level refinements. By analyzing actions in the spectral domain using the discrete cosine transform (DCT), we observe that low-frequency components capture global motion trajectories, while high-frequency components encode precise timing, alignment, and contact behaviors. Motivated by this structure, we propose Causal Spectral Policy (CSP), which models action generation as a causal coarse-to-fine process: coarse motion is predicted from observation and language, and fine corrections are generated conditionally on the realized trajectory. Across simulation and real-world evaluations, CSP consistently outperforms strong baselines on precision-sensitive manipulation tasks. Additionally, we propose human-inspired teleoperation noise injection as a data augmentation method under which our approach demonstrates strong robustness to noisy demonstrations.

Deep Learning · Attention Mechanisms

Chao Wang, Pengfei Zuo, Zhangyu Chen, Qihui Zhou, Tsung-Yi Ho, Ming-Chang Yang

Sparse attention has emerged as a vital technique for long-context inference in Large Language Models (LLMs), effectively accelerating memory-bound decoding by reducing memory access for non-essential keys. However, the assumption that decoding attention is memory-bound has been shattered. The proliferation of Multi-head Latent Attention (MLA) and Multi-Token Prediction (MTP) architectures has effectively rendered the process compute-bound. We observe that, in MLA, Q-heads exhibit a degree of sparsity even when attending to the same key; consequently, traditional sparse attention algorithms introduce significant computational inefficiency in this new regime by rigidly computing interactions between all associated Q-heads and the retrieved keys. To address this, we propose TileSparse, an arithmetic-intensity-aware (a.i.-aware) algorithm for efficient attention in compute-bound settings. We first introduce a cost model that emphasizes compute budget (compute tile size) rather than memory budget (fetched tokens) when evaluating sparse methods. Next, QK 2D Sparsity prunes unnecessary Q-head--key computations and uses the freed compute to retrieve more semantically important tokens. Because Q-head sparsity differs across keys, we further propose Tiered QK 2D Sparsity and an AutoTuner to choose the best pattern. Experiments show that under tight budgets our method improves accuracy by 40% over state-of-the-art dynamic K-only sparse methods. It also preserves 99% of full-attention accuracy while cutting attention compute by 40.8%, outperforming prior sparse attention approaches.

Deep Learning · Large Language Models

Zarif Ikram, Arad Firouzkouhi, Stephen Tu, Mahdi Soltanolkotabi, Paria Rashidinejad

A central challenge in large language model (LLM) editing is capability preservation: methods that successfully change targeted behavior can quietly game the editing proxy and corrupt general capabilities, producing degenerate behaviors reminiscent of proxy/reward hacking. We present CrispEdit, a scalable and principled second-order editing algorithm that treats capability preservation as an explicit constraint, unifying and generalizing several existing editing approaches. CrispEdit formulates editing as constrained optimization and enforces the constraint by projecting edit updates onto the low-curvature subspace of the capability-loss landscape. At the crux of CrispEdit is expressing capability constraint via Bregman divergence, whose quadratic form yields the Gauss–Newton Hessian exactly and even when the base model is not trained to convergence. We make this second-order procedure efficient at the LLM scale using Kronecker-factored approximate curvature (K-FAC) and a novel matrix-free projector that exploits Kronecker structure to avoid constructing massive projection matrices. Across standard model-editing benchmarks, CrispEdit achieves high edit success while keeping capability degradation below 1% on average across datasets, significantly improving over prior editors.

Applications · Everything Else

Zimu Lu, Houxing Ren, Yunqiao Yang, Ke Wang, Zhuofan Zong, Mingjie Zhan, Hongsheng Li

Assisting non-expert users to develop complex interactive websites has become a popular task for LLM-powered code agents. However, existing code agents tend to only generate frontend web pages, masking the lack of real full-stack data processing and storage with fancy visual effects. Notably, constructing production-level full-stack web applications is far more challenging than only generating frontend web pages, demanding careful control of data flow, comprehensive understanding of constantly updating packages and dependencies, and accurate localization of obscure bugs in the codebase. To address these difficulties, we introduce FullStack-Agent, a unified agent system for full-stack agentic coding that consists of three parts: (1) FullStack-Dev, a multi-agent framework with strong planning, code editing, codebase navigation, and bug localization abilities. (2) FullStack-Learn, an innovative data-scaling and self-improving method that back-translates crawled and synthesized website repositories to improve the backbone LLM of FullStack-Dev. (3) FullStack-Bench, a comprehensive benchmark that systematically tests the frontend, backend and database functionalities of the generated website. Our FullStack-Dev outperforms the previous state-of-the-art method by 8.7\%, 38.2\%, and 15.9\% on the frontend, backend, and database test cases respectively. Additionally, FullStack-Learn raises the performance of a 30B model by 9.7\%, 9.5\%, and 2.8\% on the three sets of test cases through self-improvement, demonstrating the effectiveness of our approach.

Deep Learning · Large Language Models

Maggie Ziyu Huan, Yuetai Li, Tuney Zheng, Xiaoyu Xu, Seungone Kim, Minxin Du, Radha Poovendran, Graham Neubig, Xiang Yue

Math reasoning has become the poster child of progress in large language models (LLMs), with new models rapidly surpassing human-level performance on benchmarks like MATH and AIME. But as math leaderboards improve week by week, it is worth asking: do these gains reflect broader problem-solving ability or just narrow overfitting? To answer this question, we evaluate over 20 open-weight reasoning-tuned models across a broad suite of tasks, including math, scientific QA, agent planning, coding, and standard instruction-following. We surprisingly find that most models that succeed in math fail to transfer their gains to other domains. To rigorously study this phenomenon, we conduct controlled experiments using math-only data with two widely-used methods: Reinforcement Learning (RL) and Supervised Finetuning (SFT) with detailed ablations. On top of the observation that RL-tuned models transfer better than SFT-tuned model, we identify on-policy fine-tuning as the key mechanism underlying cross-domain transfer, regardless of whether the training signal comes from RL or supervised learning. Latent-space representation and token-space distribution shift analyses reveal that off-policy SFT induces substantial representation and output drift, while on-policy RL preserves general-domain structure. Our results suggest a need to rethink the post-training recipes, particularly the reliance on off-policy SFT-distilled data to advance reasoning models.

Reinforcement Learning · Batch/Offline

Evgenii Opryshko, Junwei Quan, Claas Voelcker, Yilun Du, Igor Gilitschenski

Offline goal-conditioned reinforcement learning (GCRL) often struggles with long-horizon tasks, where errors in value estimation accumulate and produce unreliable policies. It is typically assumed that effective long-term planning is infeasible without specialized training. In contrast, our work demonstrates that existing GCRL policies can complete long-horizon tasks when combined with a lightweight, training-free planning wrapper. We find that standard goal-conditioned value functions encode locally consistent geometric structure sufficient for planning. Our approach, Test-Time Graph Search (TTGS), constructs a graph over the offline dataset and employs an adaptive subgoal selection strategy. To address unreliable value estimates during shortest-path search, we propose a novel mechanism that softly penalizes long-distance transitions. Our method incurs negligible computational overhead and requires no additional supervision or parameter updates. On the OGBench benchmark, TTGS unlocks latent capabilities in diverse base learners, boosting success rates on challenging locomotion tasks from near-zero to over 90\%, often matching or outperforming methods that require complex auxiliary training.

Deep Learning · Other Representation Learning

Hansen Lillemark, Benhao Huang, Fangneng Zhan, Yilun Du, T. Anderson Keller

The natural world is richly structured over space and time. Much of this structure arises from the interplay between spatial geometry and motion. However, most existing world models ignore this structure, leading to an inability to generalize in dynamic environments. In this work, we show that enforcing equivariance between an agent's representations and the world's dynamics necessarily induces an efficient, structured memory. Concretely, we introduce Flow Equivariant World Modeling, a framework in which both self-motion and external object motion are unified as one-parameter Lie-group ``flows'' acting on a latent world memory; and models are built to be equivariant with respect to these transformations. On 2D and 3D partially observed video world modeling benchmarks, we demonstrate that Flow Equivariant World Models significantly outperform comparable state-of-the-art diffusion-based and memory-augmented world modeling architectures in their ability to track and predict the locations of moving objects over long horizons. Project page: https://anonflowm.github.io/

Theory · Learning Theory

Yufeng Xie, Yunwen Lei

Asynchronous stochastic gradient descent (ASGD) is widely adopted in distributed and federated learning. In this paper, we develop a sharp generalization analysis for ASGD by leveraging the concept of on-average model stability. For convex and smooth objectives, we establish stability and excess risk bounds under minimal assumptions, removing Lipschitz continuity, bounded noise, bounded parameter or data domains, while allowing randomly partitioned data and arbitrary delays. Our bounds are optimistic and explicitly characterize the impact of worker participation, recovering the minimax-optimal rate $O(1/\sqrt{mn})$ in balanced regimes where $mn$ denotes the sample size and implying fast rates under low-noise conditions. We further extend the analysis to non-smooth objectives with Hölder-continuous gradients and to heterogeneous data settings via random ASGD, obtaining non-vacuous excess risk guarantees in both settings. Experimental results support our theoretical findings.

General Machine Learning · Methodology

Haruki Yajima, Yusuke Matsui

Tree ensembles are machine learning models with strong predictive performance and interpretability, and remain widely used for tabular data. Standard pruning methods for tree ensembles typically optimize an accuracy–compression trade-off and may change a subset of predictions, potentially compromising decision consistency. Faithful pruning methods address this issue by preserving prediction equivalence over the entire input space, but this requirement leads to lower compression ratios. We propose **PINE**, a pruning method that provides strong guarantees within an in-distribution region. PINE preserves prediction equivalence within this region and controls the region size using a single parameter $\\alpha$ via conformal calibration. Experiments on 12 public tabular datasets show that PINE improves the compression ratio by up to 30% while maintaining a comparable rate of prediction equivalence to existing faithful pruning methods. As a result, PINE achieves an improved equivalence–compression trade-off.

Applications · Chemistry, Physics, and Earth Sciences

Sum Kyun Song, Bong Gyun Shin, JaeYong Lee

Discovering governing differential equations from observational data is a fundamental challenge in scientific machine learning. Existing symbolic regression approaches rely primarily on quantitative metrics; however, real-world differential equation modeling also requires incorporating domain knowledge to ensure physical plausibility. To address this gap, we propose DoLQ, a method for discovering ordinary differential equations with LLM-based qualitative and quantitative evaluation. DoLQ employs a multi-agent architecture: a Sampler Agent proposes dynamic system candidates, a Parameter Optimizer refines equations for accuracy, and a Scientist Agent leverages an LLM to conduct both qualitative and quantitative evaluations and synthesize their results to iteratively guide the search. Experiments on multi-dimensional ordinary differential equation benchmarks demonstrate that DoLQ achieves superior performance compared to existing methods, not only attaining higher success rates but also more accurately recovering the correct symbolic terms of ground truth equations. Our code is available at https://anonymous.4open.science/r/DoLQ/README.md.

Deep Learning · Large Language Models

Naïm Es-sebbani, Esteban Marquer, Yakoub Salhi, Zied Bouraoui

Logic provides a controlled testbed for evaluating LLM-based reasoners, yet standard SAT-style benchmarks often conflate surface difficulty (length, wording, clause order) with the structural phenomena that actually determine satisfiability. We introduce a diagnostic benchmark for \emph{2-SAT} built from parameterized families of structured 2--CNF formulas, where satisfiability is characterized by the implication graph and can be tuned along interpretable axes. Our generators isolate distinct competencies and failure modes: (i) contradiction-cycle UNSAT cores with controllable size and imbalance, (ii) SAT instances with a prescribed fraction of free variables to control solution multiplicity, (iii) planted backbones that modulate propagation, (iv) late bridge clauses that couple otherwise monotone regions to probe sensitivity to ordering and revision, and (v) symmetry/duplication variants that test abstraction under renaming and redundant structure. We evaluate LLM-based reasoners on decision accuracy and assignment validity, and quantify robustness under semantics-preserving perturbations such as clause reordering, filler clauses, and variable renaming. Across models, we observe sharp performance transitions under targeted structural interventions even when surface statistics are held fixed, revealing brittleness regimes that are invisible to aggregate SAT accuracy.

Deep Learning · Attention Mechanisms

Lukas Hauzenberger, Niklas Schmidinger, Thomas Schmied, Anamaria-Roberta Hartl, David Stap, Pieter-Jan Hoedt, Sebastian Böck, Günter Klambauer, Sepp Hochreiter

There have been numerous attempts to distill quadratic attention-based LLMs into sub-quadratic linearized architectures. However, despite extensive research, such distilled models often fail to match the performance of their teacher LLMs on various downstream tasks. We set out the goal of lossless distillation, which we define in terms of tolerance-corrected Win-and-Tie rates between student and teacher on sets of tasks. We propose an additional merging stage, where individually linearized experts are combined into a single model. We show the effectiveness of this pipeline by distilling base and instruction-tuned models from the Llama, Qwen, and Olmo families. In many settings, our xLSTM-based students recover most of the teacher's performance, and even exceed it on some downstream tasks. Our contributions are an important step towards more energy-efficient and cost-effective replacements for transformer-based LLMs.

Social Aspects · Alignment

Muhammed Ustaomeroglu, Guannan Qu

Emergent misalignment can arise when a language model is fine-tuned on a narrowly scoped supervised objective: the model learns the target behavior, yet also develops undesirable out-of-domain behaviors. We investigate a mechanistic approach to preventing emergent misalignment by identifying a small set of internal features that reliably control the misaligned behavior and then discouraging the model from strengthening these features during fine-tuning. Across six fine-tuning domains, blocking (i.e., constraining) a fixed set of features achieves up to 97\% relative reduction in emergent misalignment with no degradation in target-task performance. We strengthen validity with disjoint selection/evaluation splits, multiple independent judges, multiple random seeds for key settings, quality metrics, and extensive ablations demonstrating that the reduction in misalignment is specific to the identified mechanism. We also characterize a limiting regime in which misalignment re-emerges under prolonged fine-tuning, present evidence consistent with rerouting through alternative features or layers, and evaluate modifications that partially restore the misalignment-blocking effect. Overall, our results show that targeted training-time constraints on internal mechanisms can mitigate emergent misalignment without degrading target-task performance.