论文检索

Social Aspects · Alignment

Yiran Shen, Yu Xia, Jonathan Chang, Prithviraj Ammanabrolu

Aligning large language models to human preferences is inherently multidimensional, yet most pipelines collapse heterogeneous signals into a single objective. We seek to answer what it would take to simultaneously align a model across various domains spanning those with: verifiable rewards, non-verifiable subjective preferences, and complex interactive scenarios. Such multi-objective alignment setups are often plagued by the individual objectives being at odds with each other, resulting in inefficient training and little user control during inference. To address these issues, we propose a unified framework that standardizes PRM training across verifiable and non-verifiable settings for step-level supervision, performs vectorized multi-objective alignment with Multi-Action-Head DPO, and enables controllable inference via objective-specific weighting and PRM-guided decoding. Experiments across math reasoning, human value alignment, and multi-turn tutoring show that our framework jointly improves multiple objectives simultaneously with limited interference, while remaining generalizable and adaptable across domains and offering flexible user control at inference time.

Applications · Computer Vision

Yuanyang Cao, Xichun Liu, Fuwei Zhang, Shangqi Deng, Ziyang Ren, Jianji Wang

Existing Parameter-Efficient Fine-Tuning (PEFT) methods are fundamentally constrained by a static allocation paradigm, which overlooks the model's evolving optimization priorities during training. To address this, we introduce Dynamic Adaptive Fine-tuning (DAF), a novel framework that periodically evaluates and reconfigures the trainable structure based on a context-aware decoupled sensitivity analysis. DAF employs a Rebuild-and-Refocus strategy to preserve learned knowledge by freezing outdated modules while decisively reallocating the parameter budget to newly identified critical regions. Extensive experiments on challenging vision benchmarks demonstrate that DAF significantly outperforms mainstream static PEFT methods and achieves state-of-the-art (SOTA) performance and efficiency, particularly under extreme parameter budgets. Our work fundamentally challenges the static nature of the field, offering a more intelligent and efficient paradigm for adapting large pretrained models. The code is available at https://anonymous.4open.science/r/DAF-9372.

Deep Learning · Large Language Models

Wentao Hu, Zeyu Zhu, Mingkuan Zhao, Zhenhua An, Yanbo Zhai, Shanhong yu, Huilin Zhou, Xin Lai, Xiaoyan Zhu, Jiayin Wang

Sparse Mixture-of-Experts (MoE) architectures scale model capacity efficiently but suffer from massive static parameter footprints, creating significant deployment burdens on memory-constrained hardware. Existing post-training pruning methods often rely on scalar statistics, ignoring the representational geometry of expert feature spaces. This leads to sub-optimal resource allocation across layers and the retention of redundant experts. To address this, we propose a Rank-aware Geometric Expert Pruning (RaGEP) framework to compress MoE models by analyzing the geometric properties of expert activations. First, in the inter-layer allocation stage, we introduce a Rank-aware budget allocation mechanism that adaptively assigns expert budgets based on the effective rank of layer-wise representations. Second, in the intra-layer selection stage, we propose a Spectral-Salience Pruning metric that harmonizes subspace orthogonality and activation magnitude to identify high-energy orthogonal experts. Extensive experiments across MoE models of different scales show that our method consistently outperforms state-of-the-art baselines on a diverse set of zero-shot tasks, while reducing model size and inference cost. Code is available at supplementary material.

Deep Learning · Large Language Models

Mingkuan Zhao, Xiayu Sun, Wentao Hu, Suquan Chen, Jiaxuan Li, Xiaoyan Zhu, Xin Lai, Jiayin Wang

Causal language models factorize sequence probabilities using only preceding context, leaving future information unexploited during training despite its availability in the training data. This paper introduces Regret Pre-training, a self-supervised framework grounded in the Learning Using Privileged Information (LUPI) paradigm. The framework employs a dual-view architecture in which a single model generates both a causal Student distribution and a future-conditioned Teacher distribution. The training objective augments standard language modeling with a regret loss that minimizes the KL divergence from teacher to student, transferring future-aware signals to the causal representations. We investigate two teacher configurations on the OLMoE-1B-7B architecture:LocalRegret, which extends attention by one future token, andGlobalRegret, which conditions on bidirectional context with the target position masked. Experiments on nine downstream tasks following 4 billion tokens of training demonstrate that both configurations consistently outperform the baseline. On average,GlobalRegret andLocalRegret achieve 33.9% and 32.2% accuracy respectively, surpassing the baseline's 30.2%. Most notably,GlobalRegret improves BoolQ performance by 18.1 percentage points (61.0% vs 42.9%). The framework introduces no additional parameters and requires only one extra inference-mode forward pass per training step. The source code for this paper is publicly available at https://anonymous.4open.science/r/ICML13655/ to facilitate reproducibility.

Reinforcement Learning · Online

Jianghao Wu, Daniel F Schmidt, Weiqiang Wang, Jin Ye, Jianfei Cai, Yasmeen George

Reinforcement learning with verifiable rewards (RLVR) can yield large reasoning gains from very few training instances, yet its strong sensitivity to which instances are used makes data selection a central bottleneck. Most existing selection pipelines rely on training-time optimization signals and/or require access to verifiable rewards or ground-truth answers over large candidate pools, which is costly and often infeasible in specialized domains. We study RLVR data selection in a setting where selection must be performed \emph{before} any RL training and \emph{without} labels or reward evaluation on the full pool. % We propose \textbf{SHIFT}, a one-shot, training-free selector based solely on inference-time hidden-state dynamics. For each candidate instance, SHIFT runs a single deterministic reasoning rollout and computes a \emph{reasoning-induced representation shift} (RIRS) as the start-to-end hidden-state delta. SHIFT uses the RIRS magnitude as a lightweight proxy for instance utility and enforces coverage via a quality-weighted farthest-first coreset procedure in an RIRS-augmented feature space, producing compact subsets that scale to large unlabeled pools. % Across mathematical reasoning and medical QA benchmarks under ultra-low budgets, SHIFT consistently outperforms training-free diversity and difficulty/uncertainty baselines, improving both in-domain accuracy and transfer to harder evaluation settings. Ablations show that RIRS-based coverage and quality-weighting contribute complementary gains, and analyses indicate that RIRS is not explained by simple input/output length statistics. Code will be released.

Applications · Time Series

Da Zhang, bingyu li, Zhiyuan Zhao, Hongyuan Zhang, Junyu Gao, Xuelong Li

Medical time series are central to healthcare, enabling continuous monitoring and supporting timely clinical decisions. Despite recent progress, existing methods struggle to jointly model local-global dynamics and handle nonstationarities like baseline drift, while often failing to capture latent channel interactions. To address these challenges, we propose **MedMamba**, an end-to-end architecture that integrates state space models with domain-specific inductive biases. Specifically, MedMamba first employs multi-scale convolutional embeddings to capture discriminative local morphology. Second, to mitigate nonstationarity, we introduce a tri-branch differential state space encoder that processes raw, temporal-difference, and frequency-domain views, fusing them to emphasize informative patterns while suppressing drift. Furthermore, to uncover latent channel correlations, we design a spatial graph Mamba module that learns a directed dependency structure regularized toward sparsity and acyclicity, which obviates the need for predefined graphs. Extensive experiments on five real-world datasets demonstrate that MedMamba achieves state-of-the-art performance while maintaining linear computational complexity, and ablation studies validate each component's contribution. Anonymous code is available at https://anonymous.4open.science/r/MedMamba-30DB.

Deep Learning · Foundation Models

Alesia Ivanova, Sumeet Motwani, Jack Cai, Phil Torr, Riashat Islam, Shital Shah, Christian Schroeder de Witt, Charles London

Large language models excel at short-horizon reasoning tasks, but performance drops as reasoning horizon lengths increase. Existing approaches to combat this rely on inference-time scaffolding or step-level supervision, neither of which scales easily. In this work, we introduce a scalable method to bootstrap long-horizon reasoning capabilities using only existing, abundant short-horizon data. Our approach synthetically composes simple problems into complex, multi-step dependency chains of arbitrary length. We train models on this data using outcome-only rewards under a curriculum that automatically increases in complexity, allowing RL training to be scaled much further without saturating. Empirically, our method generalizes remarkably well: curriculum training on composed 6th-grade level math problems (GSM8K) improves accuracy on longer, competition-level benchmarks (GSM-Symbolic, MATH-500, AIME). It also transfers significantly to diverse out-of-distribution ReasoningGym domains and long-context benchmarks, indicating broader generalization. Importantly, our long-horizon improvements are significantly higher than baselines even at high pass@k, showing that models can learn new reasoning paths under RL. Theoretically, we show that curriculum RL with outcome rewards could achieve an exponential improvement in sample complexity over full-horizon training, providing training signal comparable to dense supervision. h1 therefore introduces an efficient path towards scaling RL for long-horizon problems using only existing data.

Deep Learning · Large Language Models

Kai Zhang, Xiangchao Chen, Bo Liu, Tianci Xue, Zeyi Liao, Zhihan Liu, xiyao wang, Yuting Ning, Zhaorun Chen, Xiaohan Fu 等

A long-term goal of language agents is to learn and improve through their own experience, ultimately outperforming humans in complex, real-world tasks. However, training agents from experience data with reinforcement learning remains difficult in many environments, which either lack verifiable rewards (e.g., websites) or require inefficient long-horizon rollouts (e.g., multi-turn tool use). As a result, most current agents rely on supervised fine-tuning on expert data, which is challenging to scale and generalizes poorly. This limitation stems from the nature of expert demonstrations: they capture only a narrow range of scenarios, and expose the agent to limited environment diversity. We address this limitation with a middle-ground paradigm we call *early experience*: interaction data generated by the agent's own actions, where the resulting future states serve as supervision without reward signals. Within this paradigm, we study two strategies of using such data: (1) implicit world modeling, which uses collected states to ground the policy in environment dynamics; and (2) self-reflection, where the agent learns from its suboptimal actions to improve reasoning and decision-making. Evaluation across eight diverse environments and multiple model families shows that our approaches consistently improve effectiveness and out-of-domain generalization, highlighting the value of early experience. Moreover, in environments with verifiable rewards, our results provide promising signals that early experience offers a strong foundation for subsequent reinforcement learning, making it a practical bridge between imitation learning and fully experience-driven agents.

General Machine Learning · Causality

Zeyu Liu, Zheng Li, Feng Xie, Yan Zeng, Hao Zhang, Kun Zhang

Causal effect estimation is a fundamental task in many scientific fields. Selecting appropriate covariates for adjustment is crucial for obtaining unbiased causal effects. However, most existing methods either rely on learning the global causal structure, assume the absence of latent variables, or impose the pretreatment assumption-restricts covariates to those unaffected by the treatment or outcome. These assumptions are often unrealistic in real-world scenarios, and global structure learning can be computationally intensive and inefficient. To address these challenges, we first characterize the local existence boundary of adjustment sets for causal effect estimation. Based on this characterization, we develop a novel local learning method for covariate selection in nonparametric causal effect estimation. This method accommodates the presence of latent variables and eliminates the need for the pretreatment assumption. We prove that the proposed method is both sound and complete under standard assumptions. Its effectiveness is validated through extensive experiments on both synthetic and real-world datasets.

Social Aspects · Alignment

Nadav Kunievsky, James Evans

People judge interactions with large language models (LLMs) as successful when outputs match what they want, not what they type. Yet LLMs are trained to predict the next token solely from text input, not underlying intent. Because written language is an imperfect proxy for intent, and correlations between phrasing and desired outcomes can break down in training data, models that rely too heavily on surface cues may respond inconsistently to semantically equivalent prompts. This makes it essential to evaluate whether LLMs can reliably infer user intent—especially in high-stakes settings where robustness and generalization are critical. We introduce a formal framework for assessing intent comprehension in LLMs: whether a model demonstrates robust understanding of user intent by producing consistent outputs across semantically equivalent prompts while differentiating between prompts with distinct intents. Our evaluation approach is based on a variance decomposition of model responses into three components: variability due to user intent, user articulation, and model uncertainty. Models that understand what users want, and are not overly sensitive to textual cues, should attribute most output variance to intent differences, rather than articulation style. Applying this framework across diverse domains, we find that, within the five LLaMA and Gemma models we evaluate, larger models typically assign a greater share of variance to intent, indicating stronger comprehension of intent, although gains are uneven and often modest with increasing model size. These results motivate moving beyond accuracy-only benchmarks toward semantic diagnostics that directly assess whether models understand what users intend.

Deep Learning · Large Language Models

Zihao Jing, QIUHAO Zeng, Ruiyi Fang, Yan Li, Yan Sun, Boyu Wang, Pingzhao Hu

Large language models (LLMs) enable reasoning over biomolecular structures, yet existing methods remain modality-specific and typically compress structural inputs via sequence-based tokenization or fixed-length query connectors. Such architectures either omit geometric grounding required to mitigate structural hallucinations or impose inflexible modality-fusion bottlenecks that both over-compress and misallocate structural tokens, impeding generalized all-atom reasoning. We introduce **Cuttlefish**, a unified all-atom LLM that grounds language reasoning in geometric cues while scaling modality tokens with structural complexity. First, **Scaling-Aware Patching** uses an instruction-conditioned gating mechanism to generate variable-size patches over structural graphs, adaptively scaling the query-token budget with structural complexity to mitigate fixed-length connector bottlenecks. Second, **Geometry Grounding Adapter** refines these adaptive tokens via cross-attention to modality embeddings and injects the resulting modality tokens into the LLM, exposing explicit geometric cues to reduce structural hallucination. Experiments across diverse all-atom benchmarks show that Cuttlefish achieves superior performance in heterogeneous structure-grounded reasoning.

Deep Learning · Large Language Models

Kanghyun Noh, Jinheon Choi, Yulhwa Kim

Large language models (LLMs) demand substantial computational and memory resources, posing challenges for efficient deployment. Two complementary approaches have emerged to address these issues: token-adaptive layer execution, which reduces floating-point operations (FLOPs) by selectively bypassing layers, and quantization, which lowers memory footprint by reducing weight precision. However, naively integrating these techniques leads to additional accuracy degradation due to reduced redundancy in token-adaptive models. We propose QTALE (Quantization-Robust Token-Adaptive Layer Execution for LLMs), a novel framework that enables seamless integration of token-adaptive execution with quantization while preserving accuracy. Conventional token-adaptive methods reduce redundancy in two ways: (1) by limiting the diversity of training paths explored during fine-tuning, and (2) by lowering the number of parameters actively involved in inference. To overcome these limitations, QTALE introduces two key components: (1) a training strategy that ensures diverse execution paths are actively explored during fine-tuning, and (2) a post-training mechanism that allows flexible adjustment of the execution ratio at inference to reintroduce redundancy when needed. Experimental results show that QTALE enables seamless integration of token-adaptive layer execution with quantization, showing no noticeable accuracy difference, with the gap to quantization-only models kept below 0.5\% on CommonsenseQA benchmarks. By combining token-adaptive execution for FLOPs reduction and quantization for memory savings, QTALE provides an effective solution for efficient LLM deployment.

Deep Learning · Attention Mechanisms

Erwan Fagnou, Paul Caillon, Blaise Delattre, Alexandre Allauzen

Token mixing layers play a key role in how language models can learn and generate long-range dependencies. Their efficiency relies on the necessary trade-off between decoding speed and the memory requirements, along with the cache size. Considering causal generation, this paper explores new trade-offs thanks to a unified framework which separates two crucial features: (i) the direct influence of inputs on outputs in one generation step; (ii) the recurrent propagation of information through past outputs. This framework encompasses major architectures such as attention and state-space models, but also generalizes the recurrence equations by allowing each state to depend on multiple past states rather than only the immediate predecessor. By introducing structure, we design new recurrence patterns that provably achieve the desired complexity, while providing theoretical insights on their expressivity -- trading runtime for expressivity in a principled way. Empirical validation is performed on synthetic tasks, along with language modeling. Together, these results provide a unified toolkit for the understanding and design of efficient and expressive token mixers across model families.

Deep Learning · Large Language Models

Meihua Dang, Linxin Song, Honghua Zhang, Jieyu Zhao, Guy Van den Broeck, Stefano Ermon

Generations from large language models often fail to reliably conform to logical constraints such as JSON schema. Existing locally-constrained decoding (LCD) approaches enforce constraints by myopically masking out next tokens, resulting in biased sampling and degradation in downstream performance. Recent work introduces sequential Monte Carlo (SMC) methods to mitigate such sampling biases, but designing effective proposal distributions or potential functions remains a key challenge. In this work, we propose a generic approach to construct proposals and potentials for SMC sampling from $p_{\texttt{lm}}( \cdot \mid \texttt{constraint})$. First, we show that constraints specified as finite automata (FA) can be tensorized for efficient execution on GPUs, which we use to construct *globally-constrained decoding* (GCD) proposals. In addition, leveraging the fact that a tensorized FA shares the same *circuit structure* as hidden Markov models (HMM), we circuit-multiply it with an HMM to obtain the *probabilistic GCD* (P-GCD) proposal that encodes both logical and probabilistic information about the target distribution $p_{\texttt{lm}}( \cdot \mid \texttt{constraint})$. We evaluate (P-)GCD on xLAM, a widely adopted function-calling dataset, and on CommonGen, a keyword-based constrained generation benchmark. Experiments show that compared to LCD proposals, under the same SMC sampling setup, (P-)GCD achieve faster convergence to the target distribution with significantly fewer particles.

Deep Learning · Generative Models and Autoencoders

Bowen Xue, Giuseppe Guarnera, Shuang Zhao, Zahra Montazeri

Current video diffusion models generate visually compelling content but often violate basic laws of physics, producing subtle artifacts like rubber-sheet deformations and inconsistent object motion. We introduce a frequency-domain physics prior that improves motion plausibility without modifying model architectures. Our method decomposes common rigid motions (translation, rotation, scaling) into lightweight spectral losses computed on a low-frequency subset. Applied to Open-Sora, MVDIT, and Hunyuan, our approach improves both motion accuracy and action recognition by ~11\% on average on OpenVID-1M (relative), while maintaining visual quality. User studies show 74--83\% preference for our physics-enhanced videos. It also reduces warping error by 22--37\% (depending on the backbone) and improves temporal consistency scores. These results indicate that simple, global spectral cues are an effective drop-in regularizer for physically plausible motion in video diffusion.

Applications · Robotics

Tao Sun, Utkarsh Mishra, Jiaxin Lu, Danfei Xu, Iro Armeni

Compositional diffusion planners enable robotic decision-making beyond the horizon of training trajectories. Yet, current approaches often rely on the heuristic stitching of local predictions. We demonstrate that this induces a non-conservative vector field that does not mathematically correspond to any valid global trajectory log-density function. We propose Energy-based Compositional Diffuser (ECD), a framework that formulates the global trajectory as the minimizer of the sum of local bridge potentials. This energy-based perspective guarantees a conservative update field by construction and reveals a critical endpoint reaction term that is missing in heuristic stitching methods. To enable efficient inference, we further introduce a Markov-based score approximation that computes the reaction term though a single block-tridiagonal solve, maintaining time complexity linear in the planning horizon. Empirically, ECD achieves state-of-the-art success rates on a range of OGBench stitching tasks, while nearly matching the inference speed of heuristic stitching methods.

Optimization · Large Scale, Parallel and Distributed

Yihan Zhang, Xinghua Shi, Meikang Qiu, Yu Wang, Hongchang Gao

The stochastic minimax optimization problem has widespread applications in machine learning. Recently, numerous distributed minimax optimization algorithms have been developed to handle distributed training data. However, most of these algorithms suffer from high communication costs. To address this issue, we develop a novel communication-efficient decentralized stochastic gradient descent ascent with momentum algorithm based on the error feedback mechanism. Importantly, our algorithm demonstrates how to balance the full-precision update and the compression residual with novel designs for coefficients regarding variables and gradients to guarantee convergence. However, compressing the primal and dual variables (and their gradients) of stochastic minimax optimization problems with the error feedback mechanism presents significant challenges for convergence analysis. In particular, it incurs the circle dependence among consensus errors and compression errors. To overcome this challenge, we propose novel strategies that enable the establishment of the convergence rate for our algorithm. Our theoretical results demonstrate how the compression operator influences the convergence rate. Finally, extensive experimental results confirm the efficacy of our proposed algorithm.

Shih-Hsin Wang, Joel Keller, Taos Transue, Drake Brown, Thomas Strohmer, Bao Wang

Generative models that transport a simple source distribution to a complex data distribution—such as diffusion and flow-based models—are central to high‑fidelity data generation. Test-time guidance can further steer pretrained models toward user-specified high-reward regions without costly retraining. However, existing guidance methods face critical limitations: they struggle with non-differentiable rewards, fail to navigate complex landscapes, and often lack theoretical guarantees on generation performance. We propose {\it Source Parallel Tempering (SPT)}, a gradient‑free test‑time guidance framework that operates entirely in source space, leveraging its simpler geometry to avoid the complexities of the data manifold. SPT couples a local exploration kernel with parallel tempering, enabling efficient barrier crossing and robust discovery of high‑reward modes. Theoretically, we provide a new error bound linking training-time approximation error to test-time guidance performance. Empirically, SPT significantly improves over state-of-the-art methods on benchmark tasks in conditional image synthesis, protein structure generation, and dynamical system trajectory sampling.

Deep Learning · Sequential Models, Time series

Jinglin Li, Jun Tan, QI Fang, Ning Gui

Effectively modeling non-stationary dynamics in probabilistic multivariate time series(MTS) forecasting requires balancing expressiveness with robustness. Existing parametric approaches benefit from strong inductive biases but lack flexibility, whereas deep generative models struggle to capture complex temporal dependencies without extensive data and compute. We introduce Parametric Prior Mapping (PPM), a framework that injects parametric structural priors into a generative modeling process. Specifically, PPM utilizes a parametric estimator to derive a dynamic, adaptive prior, which guides the learning of a complex predictive distribution via a learnable mapping. This design allows the model to retain the efficiency of parametric methods while exploiting the expressive power of generative models. Trained with a hybrid objective, PPM yields precise forecasts with well-calibrated uncertainty estimates. Empirical results show that PPM outperforms existing baselines in handling non-stationary data, offering a superior trade-off between accuracy and computational efficiency.

Yifeng Gao, Yifan Ding, Li Wang, Feida Huang, Ye Sun, Yixu Wang, Xin Wang, Yutao Wu, Hanxun Huang, Yunhao Feng 等

The accelerating realism of AI-generated content has amplified the spread of deceptive information and eroded public trust. Prior works typically split the problem into two tracks, media authenticity, which concerns whether content is real or AI-generated, and content veracity, which concerns semantic and factual correctness, thereby missing their joint effects in practice. We present FakeWorld 1.0, which deeply fuses these two orthogonal axes into a unified omni-modal benchmark. Along the media axis, FakeWorld spans text, audio, image, and video synthesis; along the content axis, it instantiates cross-modal semantic inconsistencies and factual errors. These axes are jointly instantiated within realistic web-based and streaming-style presentation scenarios, reflecting how multimodal deception is composed and delivered in real-world settings. FakeWorld provides explainable annotations in the form of per-instance rationales, enabling transparent, evidence-based diagnosis. Under a unified protocol, our evaluation of open- and closed-source MLLMs exposes capacity limits and highlights FakeWorld’s effectiveness at surfacing mixed-source, high-fidelity deception. Beyond the benchmark, we introduce OmniCheck, a unified omni-modal agentic workflow that performs explainable detection across both axes and outputs evidence-backed reports. We aim for FakeWorld 1.0 to serve as a realistic stress test and a practical foundation for building future systems that enable scalable, explainable detection of fake multimodal content.