ICLR 2026 论文列表

We propose Any-order Any-subset Autoregressive modeling (A3), a novel sequence generation framework that generalizes standard autoregressive (AR) factorization to support the prediction of arbitrary token groups in any order. A3 overcomes the limitations of conventional left-to-right decoding by enabling flexible groupwise generation while preserving probabilistic rigor and training stability. Our design combines a two-stream attention architecture with a progressive training strategy, allowing both efficient parallel decoding and robust modeling of diverse dependency structures. Empirical results demonstrate that A3 achieves a superior trade-off between generation speed, flexibility, and quality compared to state-of-the-art AR and diffusion-based methods. This work offers a unified approach for a flexible, efficient, and novel language modeling paradigm.

The training paradigm for large language models (LLMs) is moving from static datasets to experience-based learning, where agents acquire skills via interacting with complex environments. To facilitate this transition we introduce GEM (General Experience Maker), an open-source environment simulator designed for the age of LLMs. Analogous to OpenAI-Gym for traditional reinforcement learning (RL), GEM provides a standardized framework for the environment-agent interface, including asynchronous vectorized execution for high throughput, and flexible wrappers for easy extensibility. GEM also features a diverse suite of environments, robust integrated tools, and single-file example scripts demonstrating using GEM with five popular RL training frameworks. Along with this, we also provide a set of baselines across 24 environments using REINFORCE with Return Batch Normalization (ReBN), which---unlike GRPO---is compatible with the full RL setting of dense per-turn rewards and arbitrary discount factors. We further conduct apple-to-apple benchmarking of PPO, GRPO and REINFORCE in both single- and multi-turn settings using GEM to shed light on the algorithmic designs. GEM also functions as a convenient evaluation toolkit besides a training environment. We hope this framework can help accelerate future agentic LLM research.

Model reduction, which aims to learn a simpler model of the original mixed integer linear programming (MILP), can solve large-scale MILP problems much faster. Most existing model reduction methods are based on variable reduction, which predicts a solution value for a subset of variables. From a dual perspective, constraint reduction that transforms a subset of inequality constraints into equalities can also reduce the complexity of MILP, but has been largely ignored. Therefore, this paper proposes a novel constraint-based model reduction approach for MILPs. Constraint-based MILP reduction has two challenges: 1) which inequality constraints are critical such that reducing them can accelerate MILP solving while preserving feasibility, and 2) how to predict these critical constraints efficiently. To identify critical constraints, we label the tight-constraints at the optimal solution as potential critical constraints and design an information theory-guided heuristic rule to select a subset of critical tight-constraints. Theoretical analyses indicate that our heuristic mechanism effectively identify the constraints most instrumental in reducing the solution space and uncertainty. To learn the critical tight-constraints, we propose a multi-modal representation that integrates information from both instance-level and abstract-level MILP formulations. The experimental results show that, compared to the state-of-the-art MILP solvers, our method improves the quality of the solution by over 50\% and reduces the computation time by 17.47\%.

Despite significant evolution of CUDA programming and domain-specific libraries, effectively utilizing GPUs with massively parallel engines remains difficult. Large language models (LLMs) show strong potential in generating optimized CUDA code from sequential code. However, using LLMs in practice faces two major challenges: cloud-based APIs pose risks of code leakage, and local deployment is often computationally expensive and inefficient. These drawbacks have spurred interest in small language models (SLMs), which are more lightweight and privacy-friendly. Encouragingly, recent studies show that SLMs can achieve performance comparable to LLMs on specific tasks. While SLMs can match LLMs on domain-specific tasks, their limited reasoning abilities lead to suboptimal performance in complex CUDA generation according to our experiments. To bridge this gap, we propose ReGraphT, a training-free, retrieval-augmented generation framework that transfers LLM-level reasoning to smaller models. ReGraphT organizes CUDA optimization trajectories into a structured reasoning graph, modeling the combined CUDA optimizations as state transitions, and leverages Monte Carlo Graph Search (MCGS) for efficient exploration. We also present a CUDA-specific benchmark with difficulty tiers defined by reasoning complexity to evaluate models more comprehensively. Experiments show that ReGraphT outperforms HPC-specific fine-tuned models and other retrieval-augmented approaches, achieving an average 2.33× speedup on CUDAEval and ParEval. When paired with DeepSeek-Coder-V2-Lite-Instruct and Qwen2.5-Coder-7B-Instruct, ReGraphT enables SLMs to approach LLM-level performance without the associated privacy risks or excessive computing overhead.

The design of training objective is central to training time-series forecasting models. Existing training objectives such as mean squared error mostly treat each future step as an independent, equally weighted task, which we found leading to the following two issues: (1) overlook the *label autocorrelation effect* among future steps, leading to biased training objective; (2) fail to set *heterogeneous task weights* for different forecasting tasks corresponding to varying future steps, limiting the forecasting performance. To fill this gap, we propose a novel quadratic-form weighted training objective, addressing both of the issues simultaneously. Specifically, the off-diagonal elements of the weighting matrix account for the label autocorrelation effect, whereas the non-uniform diagonals are expected to match the most preferable weights of the forecasting tasks with varying future steps. To achieve this, we propose a Quadratic Direct Forecast (QDF) learning algorithm, which trains the forecast model using the adaptively updated quadratic-form weighting matrix. Experiments show that our QDF effectively improves performance of various forecast models, achieving state-of-the-art results. Code is available at https://anonymous.4open.science/r/QDF-8937.

The open set recognition (OSR) problem aims to identify test samples from novel semantic classes that are not part of the training classes, a task that is crucial in many practical scenarios. However, the existing OSR methods use a constant scaling factor (the temperature) to the logits before applying a loss function, which hinders the model from exploring both ends of the spectrum in representation learning -- from instance-level to clas-specific features. In this paper, we address this problem by enabling temperature-modulated representation learning using a set of proposed temperature schedules, including our novel negative cosine schedule. Our temperature schedules allow the model to form a coarse decision boundary at the beginning of training by focusing on fewer neighbors, and gradually prioritizes more neighbors to smooth out the rough edges. This gradual task switching leads to a richer and more generalizable representation space. While other OSR methods benefit by including regularization or auxiliary negative samples, such as with mix-up, thereby adding a significant computational overhead, our schedules can be folded into any existing OSR loss function with no overhead. We implement the novel schedule on top of a number of baselines, using cross-entropy, contrastive and the ARPL loss functions and find that it boosts both the OSR and the closed set performance in most cases, especially on the tougher semantic shift benchmarks.

Backdoor attacks pose a significant threat to machine learning models, allowing adversaries to implant hidden triggers that alter model behavior when activated. Although gradient ascent (GA)-based unlearning has been proposed as an efficient backdoor removal approach, we identify a critical yet overlooked issue: vanilla GA does not eliminate the trigger but shifts its impact to different classes, a phenomenon we call trigger shifting. To address this, we propose Robust Gradient Ascent (RGA), which introduces a dynamic penalty mechanism to regulate GA's strength and prevent excessive unlearning. Our experiments show that RGA effectively removes backdoors while preserving model utility, offering a more reliable defense against backdoor attacks.

Deep learning models, such as wide neural networks, can be viewed as nonlinear dynamical systems composed of numerous interacting degrees of freedom. When such systems approach the limit of infinite number of degrees of freedom, their dynamics tend to simplify. This paper investigates gradient descent-based learning algorithms that exhibit linearization in their parameters. We establish that this apparent linearity, arises from weak correlations between the first, and higher-order derivatives of the hypothesis function with respect to the parameters, at initialization. Our findings indicate that these weak correlations fundamentally underpin the observed linearization phenomenon of wide neural networks. Leveraging this connection, we derive bounds on the deviation from linearity during stochastic gradient descent training. To support our analysis, we introduce a novel technique for characterizing the asymptotic behavior of random tensors. We validate our theoretical insights through empirical studies, comparing the linearized dynamics to the observed correlations.

Reinforcement learning (RL) finetuning has become a key technique for enhancing the reasoning abilities of large language models (LLMs). However, its effectiveness critically depends on the selection of training data. Recent advances underscore the importance of online prompt selection methods, which typically concentrate training on partially solved or moderately challenging examples under the current policy, thereby yielding more effective model updates. While significantly accelerating RL finetuning in terms of training steps, they also incur substantial computational overhead by requiring extensive LLM rollouts over large candidate batches to identify informative samples, an expense that can outweigh the finetuning process itself. To address this challenge, this work proposes Dynamics-Predictive Sampling (DPS), which online predicts and selects informative prompts by inferring their learning dynamics prior to costly rollouts. Specifically, we introduce a new perspective by modeling each prompt's solving progress during RL finetuning as a dynamical system, where the extent of solving is represented as the state and the transition is characterized by a hidden Markov model. Using historical rollout reward signals, we perform online Bayesian inference to estimate evolving state distributions, and the inference outcome provides a predictive prior for efficient prompt selection without rollout-intensive filtering. Empirical results across diverse reasoning tasks, including mathematics, planning, and visual geometry, demonstrate that DPS substantially reduces redundant rollouts, accelerates the training process, and achieves superior reasoning performance.

$\textbf{FAST-DIPS}$ is a training-free solver for diffusion-prior inverse problems, including nonlinear forward operators. At each noise level, a pretrained denoiser provides an anchor $\mathbf{x}_ {0|t}$; we then perform a hard-constrained proximal correction in measurement space (AWGN) by solving $\min_\mathbf{x} \tfrac{1}{2\gamma_t}\|\mathbf{x}-\mathbf{x}_{0|t}\|^2 \ \text{s.t.}\ \|\mathcal{A}(\mathbf{x})-\mathbf{y}\|\le\varepsilon$. The correction is implemented via an adjoint-free ADMM with a closed-form projection onto the Euclidean ball and a few steepest-descent updates whose step size is analytic and computable from one VJP and one JVP—or a forward-difference surrogate—followed by decoupled re-annealing. We show this step minimizes a local quadratic model (with backtracking-based descent), any ADMM fixed point satisfies KKT for the hard-constraint, and mode substitution yields a bounded time-marginal error. We also derive a latent variant $\mathcal{A}\mapsto\mathcal{A}\circ\mathcal{D}$ and a one-parameter pixel$\rightarrow$latent hybrid schedule. FAST-DIPS delivers comparable or better PSNR/SSIM/LPIPS while being substantially faster, requiring only autodiff access to $\mathcal{A}$ and no hand-coded adjoints or inner MCMC.

Test-time compute allocation in large reasoning models (LRMs) is widely used and has applications in mathematical problem solving, code synthesis, and planning. Recent work has addressed this problem by scaling self-consistency and parallel thinking, adding generic thinking tokens and prompting models to re-read the question before answering. Unfortunately, these approaches either inject task-agnostic tokens or mandate heuristics that do not explain---and often ignore---the \emph{spontaneous} repetition that many LRMs exhibit at the head of their internal chains. In contrast, we analyze and harness the model's tendency to restate the question, which we term the \emph{Echo of Prompt (EOP)}, as a front-loaded, compute-shaping mechanism. We formalize its probabilistic cost by casting echo removal as rejection-based conditioning and defining the \emph{Echo Likelihood Gap} $\Delta\mathcal{L}$ as a computable proxy. This provides the missing theoretical link that links early repetition to likelihood gains and downstream accuracy. However, it does not by itself specify how to exploit EOP. Consequently, we develop \emph{Echo-Distilled SFT (ED-SFT)} to instill an ``echo-then-reason'' pattern through supervised finetuning, and \emph{Echoic Prompting (EP)} to re-ground the model mid-trace without training. While promising, quantifying benefits beyond verbosity is non-trivial. Therefore, we conduct length and suffix-controlled likelihood analyses together with layer-wise attention studies, showing that EOP increases answer to answer-prefix attention in middle layers, consistent with an \emph{attention refocusing} mechanism. We evaluate under identical decoding settings and compute budgets on GSM8K, MathQA, Hendrycks-MATH, AIME24, and MATH-500 under identical decoding settings and budgets, and find consistent gains over baselines.

Robot task planning decomposes human instructions into executable action sequences that enable robots to complete a series of complex tasks. Although recent large language model (LLM)-based task planners achieve amazing performance, they assume that human instructions are clear and straightforward. However, real-world users are not experts, and their instructions to robots often contain significant vagueness. Linguists suggest that such vagueness frequently arises from referring expressions (REs), whose meanings depend heavily on dialogue context and environment. This vagueness is even more prevalent among the elderly and children, whom robots should serve more. This paper studies how such vagueness in REs within human instructions affects LLM-based robot task planning and how to overcome this issue. To this end, we propose the first robot task planning benchmark with vague REs (REI-Bench), where we discover that the vagueness of REs can severely degrade robot planning performance, leading to success rate drops of up to 36.9\%. We also observe that most failure cases stem from missing objects in planners. To mitigate the REs issue, we propose a simple yet effective approach: task-oriented context cognition, which generates clear instructions for robots, achieving state-of-the-art performance compared to aware prompts, chains of thought, and in-context learning. By tackling the overlooked issue of vagueness, this work contributes to the research community by advancing real-world task planning and making robots more accessible to non-expert users, e.g., the elderly and children.

The ability for AI agents to "think with images" requires a sophisticated blend of reasoning and perception. However, current open multimodal agents still largely fall short on the reasoning aspect that are crucial for real-world tasks like analyzing documents with dense charts/diagrams or navigating maps. To address this gap, we first introduce o3-bench, a new benchmark designed to evaluate multimodal reasoning while attending to visual details. O3-bench features challenging questions that require agents to gather subtle visual information from multiple distinct areas of an image while performing complex, interleaved reasoning using the gathered information. These tasks are highly challenging even for frontier systems like OpenAI o3, which only obtains 42.8% accuracy on o3-bench. To tackle these tasks, we propose InSight-o3, a multi-agent framework that divides labor between a visual reasoning agent (vReasoner) and a visual search agent (vSearcher). As a concrete first step towards o3-like systems, we focus on the latter (i.e., vSearcher) in this paper, for which we introduce the task of generalized visual search---locating relational, fuzzy, or conceptual regions described in free-form language, beyond just simple objects or figures in natural images. We present a multimodal LLM purpose-trained for this task via reinforcement learning. As a plug-and-play agent that can be directly called by other agents, our vSearcher significantly improves the performance of existing frontier multimodal models by empowering them with generalized visual search on a wide range of benchmarks.

In recent years, the research focus of large language models (LLMs) and agents has shifted increasingly from demonstrating novel capabilities to complex reasoning and tackling challenging tasks. However, existing evaluations focus mainly on math/code contests or general tasks, while existing multi-domain academic benchmarks lack sufficient reasoning depth, leaving the field without a rigorous benchmark for high-level reasoning. To fill this gap, we introduce the ACADREASON benchmark, designed to evaluate the ability of LLMs and agents to acquire and reason over academic knowledge. It consists of 50 expert-annotated academic problems across five high-reasoning domains, including computer science, economics, law, mathematics, and philosophy. All questions are sourced from top-tier publications in recent years and undergo rigorous annotation and quality control to ensure they are both challenging and answerable. We conduct systematic evaluations over 10 mainstream LLMs and agents. The results show that most LLMs scored below 20 points, with even the cutting-edge GPT-5 achieving only 16 points. While agents achieved higher scores, none exceeded 40 points. This demonstrates the current capability gap between LLMs and agents in super-intelligent academic research tasks and highlights the challenges of ACADREASON. The code and data for the ACADREASON benchmark are available at https://anonymous.4open.science/r/Acadreason-Benchmark-1BD3/.

Autoregressive models, often built on Transformer architectures, represent a powerful paradigm for generating ultra-long videos by synthesizing content in sequential chunks. However, this sequential generation process is notoriously slow. While caching strategies have proven effective for accelerating traditional video diffusion models, existing methods assume uniform denoising across all frames—an assumption that breaks down in autoregressive models where different video chunks exhibit varying similarity patterns at identical timesteps. In this paper, we present FlowCache, the first caching framework specifically designed for autoregressive video generation. Our key insight is that each video chunk should maintain independent caching policies, allowing fine-grained control over which chunks require recomputation at each timestep. We introduce a chunkwise caching strategy that dynamically adapts to the unique denoising characteristics of each chunk, complemented by an importance-based KV cache compression mechanism that maintains fixed memory bounds while preserving generation quality. Our method achieves remarkable speedups of $\textbf{2.38}\times$ on MAGI-1 and $\textbf{6.7}\times$ on SkyReels-V2, with negligible quality degradation (VBench: $0.87\uparrow$ and $0.79\downarrow$ respectively). These results demonstrate that FlowCache, successfully unlocks the potential of autoregressive models for real-time, ultra-long video generation—establishing a new benchmark for efficient video synthesis at scale. The code is available at https://anonymous.4open.science/r/FlowCache-23495iclrAnonymous

We introduce a diffusion-based framework that generates aligned novel view images and geometries via a warping‐and‐inpainting methodology. Unlike prior methods that require dense posed images or pose-embedded generative models limited to in‐domain views, our method leverages off‐the‐shelf geometry predictors to predict partial geometries viewed from reference images, and formulates novel view synthesis as an inpainting task for both image and geometry. To ensure accurate alignment between the generated image and geometry, we propose cross-modal attention instillation where the attention maps from an image diffusion branch are injected into a parallel geometry diffusion branch during both training and inference. This multi-task approach achieves synergistic effects, facilitating both geometrically robust image synthesis and geometry prediction. We further introduce proximity‐based mesh conditioning to integrate depth and normal cues, interpolating between point cloud and filtering erroneously predicted geometry from influencing the generation process. Empirically, our method achieves high-fidelity extrapolative view synthesis, delivers competitive reconstruction under interpolation settings, and produces geometrically aligned point clouds as 3D completion.

Standard attention stores keys/values losslessly but reads them via a per-head convex average, blocking channel-wise selection. We propose the Free Energy Mixer (FEM): a free-energy (log-sum-exp) read that applies a value-driven, per-channel log-linear tilt to a fast prior (e.g., from queries/keys in standard attention) over indices. Unlike methods that attempt to improve and enrich the $(q,k)$ scoring distribution, FEM treats it as a prior and yields a value-aware posterior read at unchanged complexity, smoothly moving from averaging to per-channel selection as the learnable inverse temperature increases, while still preserving parallelism and the original asymptotic complexity ($O(T^2)$ for softmax; $O(T)$ for linearizable variants). We instantiate a two-level gated FEM that is plug-and-play with standard and linear attention, linear RNNs and SSMs. It consistently outperforms strong baselines on NLP, vision, and time-series at matched parameter budgets.

Audiovisual video captioning aims to generate semantically rich descriptions with temporal alignment between visual and auditory events, thereby benefiting both video understanding and generation. In this paper, we present **AVoCaDO**, a powerful audiovisual video captioner driven by the temporal orchestration between audio and visual modalities. We propose a two-stage post-training pipeline: (1) **AVoCaDO SFT**, which fine-tunes the model on a newly curated dataset of 107K high-quality, temporally-aligned audiovisual captions; and (2) **AVoCaDO GRPO**, which leverages tailored reward functions to further enhance temporal coherence and dialogue accuracy while regularizing caption length and reducing collapse. Experimental results demonstrate that AVoCaDO significantly outperforms existing open-source models across four audiovisual video captioning benchmarks, and also achieves competitive performance on the VDC benchmark under visual-only settings. The model will be made publicly available to facilitate future research in audiovisual video understanding and generation.

Large Language Models (LLMs) have recently emerged as a powerful backbone for recommender systems. Existing LLM-based recommender systems take two different approaches for representing items in natural language, i.e., Attribute-based Representation and Description-based Representation. In this work, we aim to address the trade-off between efficiency and effectiveness that these two approaches encounter, when representing items consumed by users. Based on our observation that there is a significant information overlap between images and descriptions associated with items, we propose a novel method, **I**mage is all you need for **LLM**-based **Rec**ommender system (I-LLMRec). Our main idea is to leverage images as an alternative to lengthy textual descriptions for representing items, aiming at reducing token usage while preserving the rich semantic information of item descriptions. Through extensive experiments on real-world Amazon datasets, we demonstrate that I-LLMRec outperforms existing methods that leverage textual descriptions for representing items in both efficiency and effectiveness by leveraging images. Moreover, a further appeal of I-LLMRec is its ability to reduce sensitivity to noise in descriptions, leading to more robust recommendations.