ICLR 2026 论文列表

Amortized Bayesian inference (ABI) has emerged as a powerful simulation-based approach for estimating complex mechanistic models, offering fast posterior sampling via generative neural networks. However, extending ABI to hierarchical models, a cornerstone of modern Bayesian analysis, remains a major challenge due to the need to simulate massive data sets and estimate thousands of parameters. In this work, we build on compositional score matching (CSM), a divide-and-conquer strategy for Bayesian updating using diffusion models. To address existing stability issues of CSM in dealing with large data sets, we couple adaptive solvers with a novel, error-damping compositional estimator. Our estimator remains stable even with hundreds of thousands of data points and parameters. We validate our approach on a controlled toy example, a high-dimensional autoregressive model, and a real-world advanced microscopy application involving over 750,000 parameters.

In this paper, built upon TAPTRv2, we present TAPTRv3. TAPTRv3 improves TAPTRv2 by addressing its shortage in querying high quality features from long videos, where the target tracking points normally undergo increasing variation over time. In TAPTRv3, we propose to utilize both spatial and temporal context to bring better feature querying along the spatial and temporal dimensions for more robust tracking in long videos. For better spatial feature querying, we identify that off-the-shelf attention mechanisms struggle with point-level tasks and present Context-aware Cross-Attention (CCA). CCA introduces spatial context into the attention mechanism to enhance the quality of attention scores when querying image features. For better temporal feature querying, we introduce Visibility-aware Long-Temporal Attention (VLTA), which conducts temporal attention over all past frames while considering their corresponding visibilities. This effectively addresses the feature drifting problem in TAPTRv2 caused by its RNN-like long-term modeling. TAPTRv3 surpasses TAPTRv2 by a large margin on most of the challenging datasets and obtains state-of-the-art performance. Even when compared with methods trained on large-scale extra internal data, TAPTRv3 still demonstrates superiority.

Large language models (LLMs) are typically trained by reinforcement learning (RL) with verifiable rewards (RLVR) and supervised fine-tuning (SFT) on reasoning traces to improve their reasoning abilities. However, how these methods shape reasoning capabilities remains largely elusive. Going beyond an accuracy-based investigation of how these two components sculpt the reasoning process, this paper introduces a novel analysis framework that quantifies reasoning paths and captures their qualitative changes under each training process (with models of 1.5B, 7B, and 14B parameters on mathematical and code domains). Specifically, we investigate the reasoning process at two levels of granularity: the trajectory-level, which examines complete reasoning outputs, and the step-level, which analyzes reasoning graphs whose nodes correspond to individual reasoning steps. Notably, clustering of unique reasoning trajectories shows complementary effects: RL compresses incorrect trajectories, whereas SFT expands correct ones. Step-level analysis reveals that RL steepens (about 2.5 times), while SFT flattens (reduced to about one-third), the decay rates of node visitation frequency, degree, and betweenness centrality distributions in the reasoning graph. This indicates that RL concentrates reasoning functionality into a small subset of steps, while SFT homogenizes it across many steps. Furthermore, by evaluating the reasoning graph topologies from multiple perspectives, we delineate the shared and distinct characteristics of RL and SFT. Our work presents a novel reasoning path perspective that explains why the current best practice of two-stage training, with SFT followed by RL, is successful, and offers practical implications for data construction and more efficient learning approaches.

Large Language Model (LLM)-based planning has advanced embodied agents in long-horizon environments such as Minecraft, where acquiring latent knowledge of goal (or item) dependencies and feasible actions is critical. However, LLMs often begin with flawed priors and fail to correct them through prompting, even with feedback. We present XENON (eXpErience-based kNOwledge correctioN), an agent that algorithmically revises knowledge from experience, enabling robustness to flawed priors and sparse binary feedback. XENON integrates two mechanisms: Adaptive Dependency Graph, which corrects item dependencies using past successes, and Failure-aware Action Memory, which corrects action knowledge using past failures. Together, these components allow XENON to acquire complex dependencies despite limited guidance. Experiments across multiple Minecraft benchmarks show that XENON outperforms prior agents in both knowledge learning and long-horizon planning. Remarkably, with only a 7B open-weight LLM, XENON surpasses agents that rely on much larger proprietary models.

We study the Nearest Neighbor Search (NNS) problem in a high-dimensional setting where data originates from a low-dimensional subspace and is corrupted by Gaussian noise. Specifically, we consider a semi-random model where $n$ points from an unknown $k$-dimensional subspace of $\mathbb{R}^d$ ($k \ll d$) are perturbed by zero-mean $d$-dimensional Gaussian noise with variance $\sigma^2$ on each coordinate. Without loss of generality, we may assume the nearest neighbor is at distance $1$ from the query, and that all other points are at distance at least $1+\varepsilon$. We assume we are given only the noisy data and are required to find NN of the uncorrupted data. We prove the following results: 1. For $\sigma \in O(1/k^{1/4})$, we show that simply performing SVD denoises the data; namely, we provably recover accurate NN of uncorrupted data (Theorem 1.1). 2. For $\sigma \gg 1/k^{1/4}$, NN in uncorrupted data is not even {\bf identifiable} from the noisy data in general. This is a matching lower bound on $\sigma$ with the above result, demonstrating the necessity of this threshold for NNS (Lemma 3.1). 3. For $\sigma \gg 1/\sqrt k$, the noise magnitude ($\sigma \sqrt{d}$) is significantly exceeds the inter-point distances in the unperturbed data. Moreover, NN in noisy data is different from NN in the uncorrupted data in general. \end{enumerate} Note that (1) and (3) together imply SVD identifies correct NN in uncorrupted data even in a regime where it is different from NN in noisy data. This was not the case in existing literature (see e.g. (Abdullah et al., 2014)). Another comparison with (Abdullah et al., 2014) is that it requires $\sigma$ to be at least an inverse polynomial in the ambient dimension $d$. The proof of (1) above uses upper bounds on perturbations of singular spaces of matrices as well as concentration and spherical symmetry of Gaussians. We thus give theoretical justification for the performance of spectral methods in practice. We also provide empirical results on real datasets to corroborate our findings.

Guidance provides a simple and effective framework for posterior sampling by steering the generation process towards the desired distribution. When modeling discrete data, existing approaches mostly focus on guidance with the first-order Taylor approximation to improve the sampling efficiency. However, such an approximation is inappropriate in discrete state spaces since the approximation error could be large. A novel guidance framework for discrete data is proposed to address this problem: We derive the exact transition rate for the desired distribution given a learned discrete flow matching model, leading to guidance that only requires a single forward pass in each sampling step, significantly improving efficiency. This unified novel framework is general enough, encompassing existing guidance methods as special cases, and it can also be seamlessly applied to the masked diffusion model. We demonstrate the effectiveness of our proposed guidance on energy-guided simulations and preference alignment on text-to-image generation and multimodal understanding tasks.

Text-guided color editing in images and videos is a fundamental yet unsolved problem, requiring fine-grained manipulation of color attributes, including albedo, light source color, and ambient lighting, while preserving physical consistency in geometry, material properties, and light-matter interactions. Existing training-free approaches provide broad applicability across editing tasks but struggle with precise color control and often introduce visual inconsistency in both edited and non-edited regions. In this work, we present ColorCtrl, a training-free color editing method that leverages the attention mechanisms of modern Multi-Modal Diffusion Transformers (MM-DiT). By disentangling structure and color through targeted manipulation of attention maps and value tokens, our method enables accurate and consistent color editing, along with word-level control of attribute intensity. Our method modifies only the intended regions specified by the prompt, leaving unrelated areas untouched. Extensive experiments on both SD3 and FLUX.1-dev demonstrate that ColorCtrl outperforms existing training-free approaches and achieves state-of-the-art performances in both edit quality and consistency. Furthermore, our method surpasses strong commercial models such as FLUX.1 Kontext Max and GPT-4o Image Generation in terms of consistency. When extended to video models like CogVideoX, our approach exhibits greater advantages, particularly in maintaining temporal coherence and editing stability. Finally, our method generalizes to instruction-based editing diffusion models such as Step1X-Edit and FLUX.1 Kontext dev, further demonstrating its versatility. Here is the website.

Retrieval-augmented generation (RAG) has strong potential for producing accurate and factual outputs by combining language models (LMs) with evidence retrieved from large text corpora. However, current pipelines are limited by static chunking and flat retrieval: documents are split into short, predetermined, fixed-size chunks, embeddings are retrieved uniformly, and generation relies on whatever chunks are returned. This design brings challenges, as retrieval quality is highly sensitive to chunk size, often introduces noise from irrelevant or misleading chunks, and scales poorly to large corpora. We present SmartChunk retrieval, a query-adaptive framework for efficient and robust long-document question answering (QA). SmartChunk uses (i) a planner that predicts the optimal chunk abstraction level for each query, and (ii) a lightweight compression module that produces high-level chunk embeddings without repeated summarization. By adapting retrieval granularity on the fly, SmartChunk balances accuracy with efficiency and avoids the drawbacks of fixed strategies. Notably, our planner can reason about chunk abstractions through a novel reinforcement learning scheme, STITCH, which boosts accuracy and generalization. To reflect real-world applications, where users face diverse document types and query styles, we evaluate SmartChunk on five QA benchmarks plus one out-of-domain dataset. Across these evaluations, SmartChunk outperforms state-of-the-art RAG baselines, while reducing cost. Further analysis demonstrates strong scalability with larger corpora and consistent gains on out-of-domain datasets, highlighting its effectiveness as a general framework for adaptive retrieval.

Autonomous web agents solve complex browsing tasks, yet existing benchmarks measure only whether an agent finishes a task, ignoring whether it does so safely or in a way enterprises can trust. To integrate these agents into critical workflows, safety and trustworthiness (ST) are prerequisite conditions for adoption. We introduce \textbf{\textsc{ST-WebAgentBench}}, a configurable and extensible framework designed as a first step toward enterprise-grade evaluation. Each of its 222 tasks is paired with ST policies, concise rules that encode constraints, and is scored along six orthogonal dimensions (e.g., user consent, robustness). Beyond raw task success, we propose the \textit{Completion Under Policy} (\textit{CuP}) metric, which credits only completions that respect all applicable policies, and the \textit{Risk Ratio}, which quantifies ST breaches across dimensions. Evaluating three open state-of-the-art agents shows their average CuP is less than two-thirds of their nominal completion rate, revealing substantial safety gaps. To support growth and adaptation to new domains, ST-WebAgentBench provides modular code and extensible templates that enable new workflows to be incorporated with minimal effort, offering a practical foundation for advancing trustworthy web agents at scale.

Causal Discovery (CD) is a powerful framework for scientific inquiry. Yet, its practical adoption is hindered by a reliance on strong, often unverifiable assumptions and a lack of robust performance assessment. To address these limitations and advance empirical CD evaluation, we present **TCD-Arena**, a modularized and extendable testing kit to assess the robustness of time series CD algorithms against stepwise more severe assumption violations. For demonstration, we conduct an extensive empirical study comprising over 50 million individual CD attempts and reveal nuanced robustness profiles for 27 distinct assumption violations. Further, we investigate CD ensembles and find that they can boost general robustness, which has implications for real-world applications. With this, we strive to ultimately facilitate the development of CD methods that are reliable for a diverse range of synthetic and potentially real-world data conditions.

This paper tackles \textbf{open-ended deep research (OEDR)}, a complex challenge where AI agents must synthesize vast web-scale information into insightful reports. Current approaches are plagued by dual-fold limitations: static research pipelines that decouple planning from evidence acquisition and monolithic generation paradigms that include redundant, irrelevant evidence, suffering from hallucination issues and low citation accuracy. To address these challenges, we introduce \textbf{WebWeaver}, a novel dual-agent framework that emulates the human research process. The planner operates in a dynamic cycle, iteratively interleaving evidence acquisition with outline optimization to produce a comprehensive, citation-grounded outline linking to a memory bank of evidence. The writer then executes a hierarchical retrieval and writing process, composing the report section by section. By performing targeted retrieval of only the necessary evidence from the memory bank via citations for each part, it effectively mitigates long-context issues and citation hallucinations. Our framework establishes a new state-of-the-art across major OEDR benchmarks, including DeepResearch Bench, DeepConsult, and DeepResearchGym. These results validate our human-centric, iterative methodology, demonstrating that adaptive planning and focused synthesis are crucial for producing comprehensive, trusted, and well-structured reports.

Studies have shown that catastrophic forgetting primarily stems from the difficulty of reactivating old memories; although parameter-efficient fine-tuning can mitigate forgetting while keeping most model parameters frozen, it still falls short in fully reawakening knowledge of prior tasks. In contrast, humans can efficiently retrieve and flexibly integrate existing experiences when learning new tasks, thereby maintaining stable performance on earlier ones. During cognition, the hippocampal EC–DG–CA3–CA1 circuit engages in multiple rounds of associative recall, and its pattern-separation and memory-completion mechanisms excel at activating historical information. Inspired by this mechanism, we propose HippoTune, a latent-space iterative retrieval strategy that embeds a query–retrieve–feedback loop within each Transformer layer. Starting from the hidden state as an initial query, the model performs a few rounds of soft key–value retrieval, projects the retrieved signals back into the query, and updates it iteratively until convergence or a preset iteration limit. Theoretically, we show this process implements a Krylov-style polynomial approximation, equivalent to a differentiable second-order preconditioner, thereby deepening retrieval in a principled way. Empirically, HippoTune outperforms classical buffer-free PEFT-CL methods by 5–8% in accuracy across three vision benchmarks, while reducing training FLOPs by 50%, effectively mitigating forgetting under tight compute constraints. Code is available at: https://anonymous.4open.science/r/HippoTune-1DF2.

Graphical User Interface (GUI) agent is pivotal to advancing intelligent human-computer interaction paradigms. Constructing powerful GUI agents necessitates the large-scale annotation of high-quality user-behavior trajectory data (\textit{i.e.}, intent–trajectory pairs) for training. However, manual annotation methods and current GUI agent data mining approaches typically face three critical challenges: high construction cost, poor data quality, and low data richness. To address these issues, we propose M$^2$-Miner, the first low-cost and automated mobile GUI agent data-mining framework based on Monte Carlo Tree Search (MCTS). For better data mining efficiency and quality, we present a collaborative multi-agent framework, comprising InferAgent, OrchestraAgent, and JudgeAgent for guidance, acceleration, and evaluation. To further enhance the efficiency of mining and enrich intent diversity, we design an intent recycling strategy to extract extra valuable interaction trajectories. Additionally, a progressive model-in-the-loop training strategy is introduced to improve the success rate of data mining. Extensive experiments have demonstrated that the GUI agent fine-tuned using our mined data achieves state-of-the-art performance on several commonly used mobile GUI benchmarks. Our work will be released to facilitate the community research.

Geometric spatial reasoning forms the foundation of many applications in artificial intelligence, yet the ability of large language models (LLMs) to operate over geometric spatial information expressed in procedural code remains underexplored. In this paper, we address this gap by formalizing the \texttt{Program-to-Geometry} task, which challenges models to translate programmatic drawing code into accurate and abstract geometric reasoning. To evaluate this capability, we present \textbf{GeoGramBench}, a benchmark of 500 carefully refined problems organized by a tailored three-level taxonomy that considers geometric complexity rather than traditional mathematical reasoning complexity. Our comprehensive evaluation of 17 frontier LLMs reveals consistent and pronounced deficiencies: even the most advanced models achieve less than 50\% accuracy at the highest abstraction level. By systematically analyzing model behaviors, our study exposes key limitations in program-driven spatial reasoning and positions GeoGramBench as an important resource for benchmarking and advancing behavioral research in symbolic-to-spatial geometric reasoning.

Recent alignment methods for large language models, most notably reinforcement learning from human feedback (RLHF), often train an auxiliary reward model to minimize a loss function on binary preference data over model responses. We study a theoretical setting inspired by principle-guided methods such as Constitutional AI, in which a small set of principles (e.g., helpfulness, toxicity) act as “voters” that guide binary comparisons---such as preferring the less toxic response. We model these principles as linear directions in an embedding space of responses, a simplifying assumption motivated by the Linear Representation Hypothesis---concepts are linear directions in representation-space---a useful first-order approximation in practice. In this \emph{linear social choice model}, Ge et al. (2024) showed that an optimal linear reward model can violate Pareto optimality (PO): From the principles-as-voters lens, this means a response A can be less helpful and more toxic than B, yet still receive a higher reward. We analyze axiomatic violations in the linear social choice setting and probe the robustness of negative results under realistic assumptions. We show that added expressivity does not resolve the issue: polynomial reward models can still fail PO. We then offer a pragmatic alternative showing that when the data uniformly covers the embedding space, broad classes of loss-based rules in the limit exactly recover the axiomatic guarantees. This yields a recipe for constitutional-style alignment with provable guarantees: enforce balanced coverage \emph{via dataset design} to restore axiomatic guarantees without abandoning standard training pipelines.

Radiance of real-world scenes typically spans a much wider dynamic range than what standard cameras can capture, often leading to saturated highlights or underexposed shadows. While conventional HDR methods merge alternatively exposed frames, most approaches remain constrained to the 2D image plane, failing to model geometry and motion consistently. To address these limitations, we present HDR-NSFF, a novel framework for reconstructing dynamic HDR radiance fields from alternatively exposed monocular videos. Our method explicitly models 3D scene flow, HDR radiance, and tone mapping in a unified end-to-end pipeline. We further enhance robustness by (i) extending semantic-based optical flow with DINO features to achieve exposure-invariant motion estimation, and (ii) incorporating a generative prior as a regularizer to compensate for sparse-view and saturation-induced information loss. To enable systematic evaluation, we construct a real-world GoPro dataset with synchronized multi-exposure captures. Experiments demonstrate that HDR-NSFF achieves state-of-the-art performance in novel view and time synthesis, recovering fine radiance details and coherent dynamics even under challenging exposure variations and large motions.

Multi-step reasoning tasks like mathematical problem solving are vulnerable to cascading failures where a single incorrect step leads to complete solution breakdown. Current LLM routing methods assign entire queries to one model, treating all reasoning steps as equal. We propose TRIM (Targeted Routing in Multi-step reasoning tasks), which routes only critical steps to larger models while letting smaller models handle routine continuations. Our key insight is that targeted step-level interventions can fundamentally transform inference efficiency by confining expensive calls to precisely those steps where stronger models prevent cascading errors. TRIM operates at step-level granularity using process reward models to identify erroneous steps and makes routing decisions based on step-level uncertainty and budget constraints. We develop four routing strategies: a simple thresholding policy, two RL-trained policies (one using full sequential features, another using aggregated statistics), and a POMDP-based approach that handles uncertainty in step-level correctness estimates. On MATH-500, the thresholding policy already surpasses contemporary routing methods with $6.51$x higher cost efficiency, while RL-trained and POMDP-based policies match the strong, expensive model’s performance using $80$% fewer expensive model tokens. All methods generalize effectively across mathematical reasoning datasets, demonstrating that step-level difficulty represents fundamental characteristics of multi-step reasoning.

As large language models (LLMs) scale to billions of parameters, full-parameter fine-tuning becomes compute- and memory-prohibitive. Parameter-efficient fine-tuning (PEFT) mitigates this issue by updating only a small set of task-specific parameters while keeping the base model frozen. Among PEFT approaches, low-rank adaptation (LoRA) is widely adopted; however, it enforces a uniform rank across layers despite substantial variation in layer importance, motivating layerwise rank allocation. Recent adaptive-rank variants (e.g., AdaLoRA) allocate ranks based on importance scores, yet typically rely on instantaneous gradients that capture only local sensitivity, overlooking non-local, pathwise effects within the same layer, which yields unstable and biased scores. To address this limitation, we introduce IGU-LoRA, an adaptive-rank LoRA that (i) computes within-layer Integrated Gradients (IG) sensitivities and aggregates them into a layer-level score for rank allocation, and (ii) applies an uncertainty-aware scheme using exponential moving averages with deviation tracking to suppress noisy updates and calibrate rank selection. Theoretically, we prove an upper bound on the composite trapezoidal rule approximation error for parameter-space IG under a pathwise Hessian-Lipschitz condition, which informs the quadrature budget. Across diverse tasks and architectures, IGU-LoRA consistently outperforms strong PEFT baselines at matched parameter budgets, improving downstream accuracy and robustness. Ablations confirm the contributions of pathwise within-layer sensitivity estimates and uncertainty-aware selection to effective rank allocation.

Developing biologically plausible learning algorithms that can achieve performance comparable to error backpropagation remains a longstanding challenge. Existing approaches often compromise biological plausibility by entirely avoiding the use of spikes for error propagation or relying on both positive and negative learning signals, while the question of how spikes can represent negative values remains unresolved. To address these limitations, we introduce Bidirectional Spike-based Distillation (BSD), a novel learning algorithm that jointly trains a feedforward and a backward spiking network. We formulate learning as a transformation between two spiking representations (i.e., stimulus encoding and concept encoding) so that the feedforward network implements perception and decision-making by mapping stimuli to actions, while the backward network supports memory recall by reconstructing stimuli from concept representations. Extensive experiments on diverse benchmarks, including image recognition, image generation, and sequential regression, show that BSD achieves performance comparable to networks trained with classical error backpropagation. These findings represent a significant step toward biologically grounded, spike-driven learning in neural networks.