ICLR 2026 论文列表

Long-context video modeling is critical for multimodal large language models (MLLMs), enabling them to process movies, online video streams, and so on. Despite its advances, handling long videos remains challenging due to the difficulty in efficiently understanding the extremely long video context. This paper aims to address this issue from aspects of the model architecture, training data, training strategy, and evaluation benchmark. First, we propose a novel Hierarchical video token Compression (HiCo) method, which leverages visual redundancy in long videos to compress long video context from Clip-level to Video-level, reducing the computation significantly while preserving essential details, achieving an extreme compression ratio of approximately 1/50 with almost no performance loss. Second, we introduce a multi-stage short-to-long learning scheme, a large-scale dataset of real-world long videos named LongVid, and a challenging “Multi-Hop Needle-In- A-Video-Haystack” benchmark. Finally, we build a powerful video MLLM named VideoChat-Flash, which shows a leading performance on both mainstream long and short video benchmarks at the 2B and 7B model scales. It first gets 99.1% accuracy over 10,000 frames in NIAH among open-source models.

Search-augmented language models combine web search with Large Language Models (LLMs) to improve response groundedness and freshness. However, analyzing these systems remains challenging: existing datasets are limited in scale and narrow in scope, often constrained to static, single-turn, fact-checking questions. In this work, we introduce \textbf{Search Arena}, a crowd-sourced, large-scale, human-preference dataset of over 24,000 paired multi-turn user interactions with search-augmented LLMs. The dataset spans diverse intents and languages, and contains full system traces with around 12,000 human preference votes. Our analysis reveals that user preferences are influenced by the number of citations and types of cited sources, even when the cited content does not directly support the associated claims, uncovering a gap between perceived and actual credibility. To assess cross-setting performance, we conduct cross-arena analyses by testing search-augmented LLMs in a general purpose chat environment and conventional LLMs in search-heavy settings. We find that web search does not degrade and may even improve performance in non-search settings; however, the quality in search settings is significantly affected if solely relying on the model's parametric knowledge. We open-sourced the dataset to support future research.

LiveCodeBench (LCB) has recently become a widely adopted benchmark for evaluating large language models (LLMs) on code-generation tasks. By curating competitive programming problems, constantly adding fresh problems to the set, and filtering them by release dates, LCB provides contamination-aware evaluation and offers a holistic view of coding capability. However, LCB remains restricted to Python, leaving open the question of whether LLMs can generalize across the diverse programming languages required in real-world software engineering. We introduce Multi-LCB, a benchmark for evaluating LLMs across twelve programming languages, including Python. Multi-LCB transforms Python tasks from the LCB dataset into equivalent tasks in other languages while preserving LCB’s contamination controls and evaluation protocol. Because it is fully compatible with the original LCB format, Multi-LCB will automatically track future LCB updates, enabling systematic assessment of cross-language code generation competence and requiring models to sustain performance well beyond Python. We evaluated 20 LLMs for instruction and reasoning on Multi-LCB, uncovering evidence of Python overfitting, language-specific contamination, and substantial disparities in multilingual performance. Our results establish Multi-LCB as a rigorous new benchmark for multi-programming-language code evaluation, directly addressing LCB’s primary limitation and exposing critical gaps in current LLM capabilities. All prompts, source code and experimental configurations are publicly available at https://anonymous.4open.science/r/Multi-LiveCodeBench-C627/

Trajectory data records the spatio-temporal movements of people and vehicles. However, raw trajectories are often noisy, incomplete, or inconsistent due to sensor errors and transmission failures. To ensure reliable downstream analytics, Trajectory Data Preparation (TDP) has emerged as a critical preprocessing stage, encompassing various tasks such as imputation, map matching, anomaly detection, trajectory recovery, compression, etc. However, existing TDP methods face two major limitations: (i) they assume centralized access to data, which is unrealistic under strict privacy regulations and data silo situations, and (ii) they train task-specific models that lack generalization across diverse or unseen TDP tasks. To this end, we propose FedTDP for Federated Trajectory Data Preparation (F-TDP), where trajectories are vertically partitioned across regions and cannot be directly shared. FedTDP introduces three innovations: (i) lightweight Trajectory Privacy AutoEncoder (TPA) with secret-sharing aggregation, providing formal privacy guarantees; (ii) Trajectory Knowledge Enhancer (TKE) that adapts LLMs to spatio-temporal patterns via trajectory-aware prompts, offsite-tuning, sparse-tuning, and bidirectional knowledge distillation; and (iii) Federated Parallel Optimization (FPO), which reduces communication overhead and accelerates federated training. We conduct experiments on 6 real-world datasets and 10 representative TDP tasks, showing that FedTDP surpasses 13 state-of-the-art baselines in accuracy, efficiency, and scalability, while also generalizing effectively across diverse TDP tasks.

Achieving political consensus is crucial yet challenging for the effective functioning of social governance. However, although frontier AI systems represented by large language models (LLMs) have developed rapidly in recent years, their capabilities in this scope are still understudied. In this paper, we introduce PoliCon, a novel benchmark constructed from 2,225 high-quality deliberation records of the European Parliament over 13 years, ranging from 2009 to 2022, to evaluate the ability of LLMs to draft consensus resolutions based on divergent party positions under varying collective decision-making contexts and political requirements. Specifically, PoliCon incorporates four factors to build each task environment for finding different political consensus: specific political issues, political goals, participating parties, and power structures based on seat distribution. We also developed an evaluation framework based on social choice theory for PoliCon, which simulates the real voting outcomes of different political parties to assess whether LLM-generated resolutions meet the requirements of the predetermined political consensus. Our experimental results demonstrate that even state-of-the-art models remain undersatisfied with complex tasks like passing resolutions by a two-thirds majority and addressing security issues, while uncovering their inherent partisan biases and revealing some behaviors LLMs show to achieve the consensus, such as prioritizing the stance of the dominant party instead of uniting smaller parties, which highlights PoliCon's promise as an effective platform for studying LLMs' ability to promote political consensus. The code and dataset are released at [PoliCon Website](https://zowiezhang.github.io/projects/PoliCon).

Reliable causal inference is essential for making decisions in high-stakes areas like medicine, economics, and public policy. However, it remains unclear whether large language models (LLMs) can handle rigorous and trustworthy \textit{statistical causal inference}. Current benchmarks usually involve simplified tasks. For example, these tasks might only ask LLMs to identify semantic causal relationships or draw conclusions directly from raw data. As a result, models may overlook important statistical pitfalls, such as Simpson’s paradox or selection bias. This oversight limits the applicability of LLMs in the real world. To address these limitations, we propose \textbf{CausalPitfalls}, a comprehensive benchmark designed to rigorously evaluate the capability of LLMs in overcoming common causal inference pitfalls. Our benchmark features structured challenges across multiple difficulty levels, each paired with grading rubrics. This approach allows us to quantitatively measure both causal reasoning capabilities and the reliability of LLMs' responses. We evaluate models using two protocols: (1) direct prompting, which assesses intrinsic causal reasoning, and (2) code-assisted prompting, where models generate executable code for explicit statistical analysis. Additionally, we validate the effectiveness of this judge by comparing its scoring with assessments from human experts. Our results reveal significant limitations in current LLMs when performing statistical causal inference. The CausalPitfalls benchmark provides essential guidance and quantitative metrics to advance the development of trustworthy causal reasoning systems.

Backdoor attacks pose severe security threats to large language models (LLMs), where a model behaves normally under benign inputs but produces malicious outputs when a hidden trigger appears. Existing backdoor removal methods typically assume prior knowledge of triggers, access to a clean reference model, or rely on aggressive finetuning configurations, and are often limited to classification tasks. However, such assumptions fall apart in real-world generative LLM settings. In this work, we propose a new framework for purifying **generative LLM** without any prior trigger knowledge or clean references. Through systematic sanity checks, we find that backdoor associations are redundantly encoded across MLP layers, while attention modules primarily amplify trigger signals without establishing the behavior. Leveraging this insight, we shift the focus from isolating specific backdoor triggers to cutting off the trigger–behavior associations, and design an immunization-inspired elimination approach: by constructing multiple synthetic backdoored variants of the given suspicious model, each trained with different malicious trigger–behavior pairs, and contrasting them with their clean counterparts. The recurring modifications across variants reveal a shared **"backdoor signature"**—analogous to antigens in a virus. Guided by this signature, we neutralize highly suspicious components in LLM and apply lightweight finetuning to restore its fluency, producing purified models that withstand diverse backdoor attacks and threat models while preserving generative capability.

Large Language Models (LLMs) have shown remarkable success, and their multimodal expansions (MLLMs) further unlock capabilities spanning images, videos, and other modalities beyond text. However, despite this shift, prompt optimization approaches, designed to reduce the burden of manual prompt crafting while maximizing performance, remain confined to text, ultimately limiting the full potential of MLLMs. Motivated by this gap, we introduce the new problem of multimodal prompt optimization, which expands the prior definition of prompt optimization to the multimodal space defined by the pairs of textual and non-textual prompts. To tackle this problem, we then propose the Multimodal Prompt Optimizer (MPO), a unified framework that not only performs the joint optimization of multimodal prompts through alignment-preserving updates but also guides the selection process of candidate prompts by leveraging earlier evaluations as priors in a Bayesian-based selection strategy. Through extensive experiments across diverse modalities that go beyond text, such as images, videos, and even molecules, we demonstrate that MPO outperforms leading text-only optimization methods, establishing multimodal prompt optimization as a crucial step to realizing the potential of MLLMs.

While machine-generated texts (MGTs) offer great convenience, they also pose risks such as disinformation and phishing, highlighting the need for reliable detection. Metric-based methods, which extract statistically distinguishable features of MGTs, are often more practical than complex model-based methods that are prone to overfitting. Given their diverse designs, we first place representative metric-based methods within a unified framework, enabling a clear assessment of their advantages and limitations. Our analysis identifies a core challenge across these methods: the token-level detection score is easily biased by the inherent randomness of the MGTs generation process. To address this, we theoretically and empirically reveal two relationships of context detection scores that may aid calibration: Neighbor Similarity and Initial Instability. We then propose a Markov-informed score calibration strategy that models these relationships using Markov random fields, and implements it as a lightweight component via a mean-field approximation, allowing our method to be seamlessly integrated into existing detectors. Extensive experiments in various real-world scenarios, such as cross-LLM and paraphrasing attacks, demonstrate significant gains over baselines with negligible computational overhead. The code is available at \url{https://anonymous.4open.science/r/MRF-Enhance}.

Parameter-Efficient Fine-Tuning (PEFT) methods have become crucial for rapidly adapting large language models (LLMs) to downstream tasks. Prefix-Tuning, an early and effective PEFT technique, demonstrated the ability to achieve performance comparable to full fine-tuning with significantly reduced computational and memory overhead. However, despite its earlier success, its effectiveness in training modern state-of-the-art LLMs has been very limited. In this work, we demonstrate empirically that Prefix-Tuning underperforms on LLMs because of an inherent tradeoff between the contribution of input prompt and parameterized prefix within the attention head. This motivates us to introduce PrefixMemory-Tuning, an architecture that generalizes the principles of Prefix-Tuning while addressing its shortcomings by shifting the prefix module out of the attention head itself and improving its expressiveness. Our experiments show that, across diverse benchmarks, PrefixMemory-Tuning consistently outperforms existing Prefix-Tuning methods. Notably, it achieves competitive performance with modern PEFTs on several general benchmarks, highlighting a potential extension of Prefix-Tuning approaches to become state-of-the-art. Our findings suggest that by overcoming its inherent limitations, Prefix-Tuning can remain a competitive and relevant research direction in the landscape of parameter-efficient LLM adaptation.

Large vision-language models (VLMs) achieve strong performance in Visual Question Answering but still rely heavily on supervised fine-tuning (SFT) with massive labeled datasets, which is costly due to human annotations. Crucially, real-world datasets often exhibit *human uncertainty* (**HU**) — variation in human confidence across annotations, but standard SFT simply optimizes toward the most frequent label, disregarding HU distributions. This leaves two open questions: *How does HU affect SFT*, and *how can HU be effectively leveraged in training?* In this work, we first conduct a systematic evaluation of VLMs across varying HU levels. We have two key findings: (i) surprisingly, high-HU samples contribute little, or even degrade, model performance, and (ii) naively training on the full dataset yields under-calibrated models that fail to capture HU distributions. Motivated by these findings, we introduce **HaDola**, a **h**uman uncertainty-**a**ware **d**ata selection and aut**o**matic **la**beling framework. HaDola operates in four stages: **discriminate**, **self-annotate**, **error trigger**, and **training**, to iteratively identify harmful samples, prioritize informative ones, and bootstrap from a small seed set (5% of data). Our approach substantially reduces reliance on costly HU annotations and makes VLMs more accurate and better calibrated. Extensive experiments on VQAv2 and VizWiz datasets demonstrate that HaDola consistently matches or outperforms state-of-the-art baselines, with less training data. Our work highlights the importance of explicitly modeling HU in SFT, suggesting better utilization of HU is more effective than merely scaling up dataset size.

The Mixture of Experts (MoE) is an effective architecture for scaling large language models by leveraging sparse expert activation to balance performance and efficiency. However, under expert parallelism, MoE suffers from inference inefficiencies due to imbalanced token-to-expert assignment, where underloaded experts complete computations early but must wait for overloaded experts, leading to global delays. We define this phenomenon as the \textbf{\textit{Straggler Effect}}, as the most burdened experts dictate the overall inference latency. To address this, we first propose \textit{\textbf{Capacity-Aware Token Drop}}, which enforces expert capacity limits by discarding excess tokens from overloaded experts, effectively reducing load imbalance with minimal performance impact (e.g., $30\%$ speedup with only $0.9\%$ degradation on OLMoE). Next, given the presence of low-load experts remaining well below the capacity threshold, we introduce \textit{\textbf{Capacity-Aware Expanded Drop}}, which allows tokens to include additional local experts in their candidate set before enforcing strict local capacity constraints, thereby improving load balance and enhancing the utilization of underused experts. Extensive experiments on both language and multimodal MoE models demonstrate the effectiveness of our approach, yielding substantial gains in expert utilization, model performance, and inference efficiency, e.g., applying Expanded Drop to Mixtral-8$\times$7B-Instruct yields a {0.2\%} average performance improvement and a {1.85$\times$} inference speedup. The code is released at: https://github.com/CASE-Lab-UMD/Capacity-Aware-MoE.

Visual Question Answering (VQA) enables targeted and context-dependent analysis of medical images, such as chest X-rays (CXRs). However, existing VQA datasets for CXRs are typically constrained by simplistic and brief answer formats, lacking localization annotations (e.g., bounding boxes) and structured tags (e.g., region or radiological finding/disease tags). To address these limitations, we introduce MIMIC-Ext-CXR-QBA (abbr. CXR-QBA), a large-scale CXR VQA dataset derived from MIMIC-CXR, comprising 42 million QA-pairs with multi-granular, multi-part answers, detailed bounding boxes, and structured tags. We automatically generated our VQA dataset from scene graphs (also made available), which we constructed using LLM-based information extraction from radiology reports. After automatic quality assessment, we identified 31M pre-training and 7.5M fine-tuning grade QA-pairs, providing the largest and most sophisticated VQA dataset for CXRs to date. Tools for using our dataset and the construction pipeline are available at https://github.com/philip-mueller/mimic-ext-cxr-qba/ .

Speech brain–computer interfaces (BCIs) aim to restore communication for people with paralysis by translating neural activity into text. Most systems use cascaded frameworks that decode phonemes before assembling sentences with an n-gram language model (LM), preventing joint optimization of all stages simultaneously. Here, we introduce an end-to-end BraIn-to-Text (BIT) framework that translates neural activity into coherent sentences using a single differentiable neural network. Central to our approach is a cross-task, cross-species pretrained neural encoder, whose representations transfer to both attempted and imagined speech. In a cascaded setting with an n-gram LM, the pretrained encoder establishes a new state-of-the-art (SOTA) on the Brain-to-Text ’24 and ’25 benchmarks. Integrated end-to-end with audio large language models (LLMs) and trained with contrastive learning for cross-modal alignment, BIT reduces the word error rate (WER) of the prior end-to-end method from 24.69% to 10.22%. Notably, we find that small-scale audio-LLMs markedly improve end-to-end decoding. Beyond record-setting performance, BIT aligns attempted and imagined speech embeddings to enable cross-task generalization. Altogether, our approach advances the integration of large, diverse neural datasets, paving the way for an end-to-end decoding framework that supports seamless, differentiable optimization.

Do large language models (LLMs) represent concepts abstractly, i.e., independent of input format? We revisit Function Vectors (FVs), compact representations of in-context learning (ICL) tasks that causally drive task performance. Across multiple LLMs, we show that FVs are not fully invariant: FVs are nearly orthogonal when extracted from different input formats (e.g., open-ended vs. multiple-choice), even if both target the same concept. We identify Concept Vectors (CVs), which carry more stable concept representations. Like FVs, CVs are composed of attention head outputs; however, unlike FVs, the constituent heads are selected using Representational Similarity Analysis (RSA) based on whether they encode concepts consistently across input formats. While these heads emerge in similar layers to FV-related heads, the two sets are largely distinct, suggesting different underlying mechanisms. Steering experiments reveal that FVs excel in-distribution, when extraction and application formats match (e.g., both open-ended in English), while CVs generalize better out-of-distribution across both question types (open-ended vs. multiple-choice) and languages. Our results show that LLMs do contain abstract concept representations, but these differ from those that drive ICL performance.

Vision-language models (VLMs) have demonstrated impressive generalization across multimodal tasks, yet most evaluation benchmarks remain Western-centric, leaving open questions about their performance in culturally diverse and multilingual settings. To address this gap, we introduce IndicVisionBench, the first large-scale benchmark centered on the Indian subcontinent. Covering English and 10 Indian languages, our benchmark spans 3 multimodal tasks, including Optical Character Recognition (OCR), Multimodal Machine Translation (MMT), and Visual Question Answering (VQA), covering 6 kinds of question types. Our final benchmark consists of a total of ~5K images and 37K+ QA pairs across 13 culturally grounded topics. In addition, we release a paired parallel corpus of annotations across 10 Indic languages, creating a unique resource for analyzing cultural and linguistic biases in VLMs. We evaluate a broad spectrum of 8 models, from proprietary closed-source systems to open-weights medium and large-scale models. Our experiments reveal substantial performance gaps, underscoring the limitations of current VLMs in culturally diverse contexts. By centering cultural diversity and multilinguality, IndicVisionBench establishes a reproducible evaluation framework that paves the way for more inclusive multimodal research.

Counterfactual reasoning, a hallmark of intelligence, consists of three steps: inferring latent variables from observations (abduction), constructing alternative situations (interventions), and predicting the outcomes of the alternatives (prediction). This skill is essential for advancing LLMs' causal understanding and expanding their applications in high-stakes domains such as scientific research and healthcare. However, existing efforts in assessing LLM's counterfactual reasoning capabilities tend to skip the abduction step, effectively reducing to interventional reasoning and leading to over-estimated LLM performance. To address this, we introduce executable counterfactuals, a novel framework that operationalizes causal reasoning through code and math problems. Our framework explicitly requires all three steps of counterfactual reasoning and enables scalable synthetic data creation with varying difficulty, creating a new frontier for evaluating and improving LLM's reasoning. Our results reveal substantial drop in accuracy (25-40%) from interventional to counterfactual reasoning for state-of-the-art models such as o4-mini and Claude-4-Sonnet. To address this gap, we construct a training set comprising counterfactual code problems having if-condition and test on out-of-distribution code structures (e.g., having while-loop); we also test whether a model trained on code would generalize to counterfactual math word problems. While Supervised Finetuning (SFT) on stronger models' reasoning traces improves in-distribution performance of Qwen models, it leads to a decrease in accuracy on out-of-distribution tasks such as counterfactual math problems. In contrast, reinforcement learning (RL) induces the core cognitive behaviors and generalizes to new distributions, yielding substantial accuracy gains over the base model on both code (improvement of 1.5x-2x) and counterfactual math problems. Analysis of the reasoning traces further reinforces these findings and highlights the promise of RL with scalable data generation for improving LLMs' counterfactual reasoning.

Scientific discovery is increasingly constrained by costly experiments and limited budgets, making efficient optimization essential for AI for science. Bayesian Optimization (BO), while widely adopted for balancing exploration and exploitation, suffers from slow cold-start performance and poor scalability in high-dimensional settings, limiting its effectiveness in real-world scientific applications. To address these challenges, we propose LLM-Guided Bayesian Optimization (LGBO), the first LLM preference-guided BO framework that continuously integrates the semantic reasoning of large language models (LLMs) into the optimization loop. Unlike prior works that use LLMs only for warm-start initialization or candidate generation, LGBO introduces a region-lifted preference mechanism that embeds LLM-driven preferences into every iteration, shifting the surrogate mean in a stable and controllable way. Theoretically, we prove that LGBO is not perform significantly worse than standard BO in the worst case, while achieving significantly faster convergence when preferences align with the objective. Empirically, LGBO achieves consistent improvements across diverse dry benchmarks in physics, chemistry, biology, and materials science. Most notably, in a new wet-lab optimization of Fe–Cr battery electrolytes, LGBO reaches \textbf{90\% of the best observed value within 6 iterations}, whereas standard BO and existing LLM-augmented baselines require more than 10 iterations. Together, the results suggest that LGBO offers a promising direction for integrating LLMs into scientific optimization workflows.

Computer-Aided Design (CAD), renowned for its flexibility and precision, serves as the foundation of digital design. Recently, some efforts adopt Large Language Models (LLMs) for generating parametric CAD command sequences from text instructions. However, our study reveals that LLMs pre-trained on large-scale general data are not proficient at directly outputting task-specific CAD sequences. Instead of relying on direct generation, we introduce a Plan then Act process where user instructions are first parsed into a chain-like operational plan via an LLM, which is then used to generate accurate command sequences. Specifically, we propose PTA, a new bi-level CAD command sequence generation method. The PTA consists of two critical stages: high-level plan generation and low-level command generation. During the high-level stage, an LLM-based Planner completes the planning process, parsing user instructions into a high-level operation plan. Following this, at the low-level generation stage, we introduce an Actioner equipped with a requirement-aware mechanism to extract design requirements (e.g., dimensions, geometric relationships) from user instructions. This extracted information is used to guide the low-level command sequence generation, improving the alignment of the generated sequences with user requirements. Experimental results demonstrate that our PTA outperforms existing methods in both quantitative and qualitative evaluations. Our source code will be made publicly available.