NeurIPS 2025 论文列表

LLM‑based agents have demonstrated great potential in generating and managing code within complex codebases. In this paper, we introduce WebGen-Bench, a novel benchmark designed to measure an LLM-based agent's ability to create multi-file website codebases from scratch. It contains diverse instructions for website generation, created through the combined efforts of human annotators and GPT-4o. These instructions span three major categories and thirteen minor categories, encompassing nearly all important types of web applications. To assess the quality of the generated websites, we generate test cases targeting each functionality described in the instructions. These test cases are then manually filtered, refined, and organized to ensure accuracy, resulting in a total of 647 test cases. Each test case specifies an operation to be performed on the website and the expected outcome of the operation. To automate testing and improve reproducibility, we employ a powerful web-navigation agent to execute test cases on the generated websites and determine whether the observed responses align with the expected results. We evaluate three high-performance code-agent frameworks—Bolt.diy, OpenHands, and Aider—using multiple proprietary and open-source LLMs as engines. The best-performing combination, Bolt.diy powered by DeepSeek-R1, achieves only 27.8\% accuracy on the test cases, highlighting the challenging nature of our benchmark. Additionally, we construct WebGen-Instruct, a training set consisting of 6,667 website-generation instructions. Training Qwen2.5-Coder-32B-Instruct on Bolt.diy trajectories generated from a subset of the training set achieves an accuracy of 38.2\%, surpassing the performance of the best proprietary model. We release our data-generation, training, and testing code, along with both the datasets and model weights at https://github.com/mnluzimu/WebGen-Bench.

Large language models (LLMs) are increasingly capable of processing long inputs and locating specific information within them, as evidenced by their performance on the Needle in a Haystack (NIAH) test. However, while models excel at recalling surprising information, they still struggle to identify clearly omitted information. We introduce AbsenceBench to assesses LLMs' capacity to detect missing information across three domains: numerical sequences, poetry, and GitHub pull requests. AbsenceBench asks models to identify which pieces of a document were deliberately removed, given access to both the original and edited contexts. Despite the apparent straightforwardness of these tasks, our experiments reveal that even state-of-the-art models like Claude-3.7-Sonnet achieve only 69.6% F1-score with a modest average context length of 5K tokens. Our analysis suggests this poor performance stems from a fundamental limitation: Transformer attention mechanisms cannot easily attend to "gaps" in documents since these absences don't correspond to any specific keys that can be attended to. Overall, our results and analysis provide a case study of the close proximity of tasks where models are already superhuman (NIAH) and tasks where models breakdown unexpectedly (AbsenceBench).

Learning editable high-resolution scene representations for dynamic scenes is an open problem with applications across the domains from autonomous driving to creative editing - the most successful approaches today make a trade-off between editability and supporting scene complexity: neural atlases represent dynamic scenes as two deforming image layers, foreground and background, which are editable in 2D, but break down when multiple objects occlude and interact. In contrast, scene graph models make use of annotated data such as masks and bounding boxes from autonomous-driving datasets to capture complex 3D spatial relationships, but their implicit volumetric node representations are challenging to edit view-consistently. We propose Neural Atlas Graphs (NAGs), a hybrid high-resolution scene representation, where every graph node is a view-dependent neural atlas, facilitating both 2D appearance editing and 3D ordering and positioning of scene elements. Fit at test-time, NAGs achieve state-of-the-art quantitative results on the Waymo Open Dataset - by 5 dB PSNR increase compared to existing methods - and make environmental editing possible in high resolution and visual quality - creating counterfactual driving scenarios with new backgrounds and edited vehicle appearance. We find that the method also generalizes beyond driving scenes and compares favorably - by more than 7 dB in PSNR - to recent matting and video editing baselines on the DAVIS video dataset with a diverse set of human and animal-centric scenes. Project Page: https://princeton-computational-imaging.github.io/nag/

Machine learning (ML) is transforming modeling and control in the physical, engineering, and biological sciences. However, rapid development has outpaced the creation of standardized, objective benchmarks—leading to weak baselines, reporting bias, and inconsistent evaluations across methods. This undermines reproducibility, misguides resource allocation, and obscures scientific progress. To address this, we propose a Common Task Framework (CTF) for scientific machine learning. The CTF features a curated set of datasets and task-specific metrics spanning forecasting, state reconstruction, and generalization under realistic constraints, including noise and limited data. Inspired by the success of CTFs in fields like natural language processing and computer vision, our framework provides a structured, rigorous foundation for head-to-head evaluation of diverse algorithms. As a first step, we benchmark methods on two canonical nonlinear systems: Kuramoto-Sivashinsky and Lorenz. These results illustrate the utility of the CTF in revealing method strengths, limitations, and suitability for specific classes of problems and diverse objectives. Next, we are launching a competition around a global real world sea surface temperature dataset with a true holdout dataset to foster community engagement. Our long-term vision is to replace ad hoc comparisons with standardized evaluations on hidden test sets that raise the bar for rigor and reproducibility in scientific ML.

Human communication is $\textit{motivated}$: people speak, write, and create content with a particular communicative intent in mind. As a result, information that large language models (LLMs) and AI agents process is inherently framed by humans' intentions and incentives. People are adept at navigating such nuanced information: we routinely identify benevolent or self-serving motives in order to decide what statements to trust. For LLMs to be effective in the real world, they too must critically evaluate content by factoring in the motivations of the source---for instance, weighing the credibility of claims made in a sales pitch. In this paper, we undertake a comprehensive study of whether LLMs have this capacity for $\textit{motivational vigilance}$. We first employ controlled experiments from cognitive science to verify that LLMs' behavior is consistent with rational models of learning from motivated testimony, and find they successfully discount information from biased sources in a human-like manner. We then extend our evaluation to sponsored online adverts, a more naturalistic reflection of LLM agents' information ecosystems. In these settings, we find that LLMs' inferences do not track the rational models' predictions nearly as closely---partly due to additional information that distracts them from vigilance-relevant considerations. However, a simple steering intervention that boosts the salience of intentions and incentives substantially increases the correspondence between LLMs and the rational model. These results suggest that LLMs possess a basic sensitivity to the motivations of others, but generalizing to novel real-world settings will require further improvements to these models.

AI agents have the potential to significantly alter the cybersecurity landscape. Here, we introduce the first framework to capture offensive and defensive cyber-capabilities in evolving real-world systems. Instantiating this framework with BountyBench, we set up 25 systems with complex, real-world codebases. To capture the vulnerability lifecycle, we define three task types: Detect (detecting a new vulnerability), Exploit (exploiting a given vulnerability), and Patch (patching a given vulnerability). For Detect, we construct a new success indicator, which is general across vulnerability types and provides localized evaluation. We manually set up the environment for each system, including installing packages, setting up server(s), and hydrating database(s). We add 40 bug bounties, which are vulnerabilities with monetary awards from \\$10 to \\$30,485, covering 9 of the OWASP Top 10 Risks. To modulate task difficulty, we devise a new strategy based on information to guide detection, interpolating from identifying a zero day to exploiting a given vulnerability. We evaluate 10 agents: Claude Code, OpenAI Codex CLI with o3-high and o4-mini, and custom agents with o3-high, GPT-4.1, Gemini 2.5 Pro Preview, Claude 3.7 Sonnet Thinking, Qwen3 235B A22B, Llama 4 Maverick, and DeepSeek-R1. Given up to three attempts, the top-performing agents are OpenAI Codex CLI: o3-high (12.5% on Detect, mapping to \\$3,720; 90% on Patch, mapping to \\$14,152), Custom Agent with Claude 3.7 Sonnet Thinking (67.5% on Exploit), and OpenAI Codex CLI: o4-mini (90% on Patch, mapping to \\$14,422). OpenAI Codex CLI: o3-high, OpenAI Codex CLI: o4-mini, and Claude Code are more capable at defense, achieving higher Patch scores of 90%, 90%, and 87.5%, compared to Exploit scores of 47.5%, 32.5%, and 57.5% respectively; while the custom agents are relatively balanced between offense and defense, achieving Exploit scores of 17.5-67.5% and Patch scores of 25-60%.

In off-policy policy evaluation (OPE) tasks within reinforcement learning, Temporal Difference Learning(TD) and Fitted Q-Iteration (FQI) have traditionally been viewed as differing in the number of updates toward the target value function: TD makes one update, FQI makes an infinite number, and Partial Fitted Q-Iteration (PFQI) performs a finite number. We show that this view is not accurate, and provide a new mathematical perspective under linear value function approximation that unifies these methods as a single iterative method solving same linear system, but using different matrix splitting schemes and preconditioners. We show that increasing the number of updates under the same target value function, i.e., the target network technique, is a transition from using a constant preconditioner to using a data-feature adaptive preconditioner. This elucidates, for the first time, why TD convergence does not necessarily imply FQI convergence, and establishes tight convergence connections among TD, PFQI, and FQI. Our framework enables sharper theoretical results than previous work and characterization of the convergence conditions for each algorithm, without relying on assumptions about the features (e.g., linear independence). We also provide an encoder-decoder perspective to better understand TD’s convergence conditions, and prove, for the first time, that when a large learning rate doesn’t work, trying a smaller one may help(for batch TD). Our framework also leads to the discovery of new crucial conditions on features for convergence, and shows how common assumptions about features influence convergence, e.g., the assumption of linearly independent features can be dropped without compromising the convergence guarantees of stochastic TD in the on-policy setting. This paper is also the first to introduce matrix splitting into the convergence analysis of these algorithms.

Large Language Models (LLMs) have demonstrated impressive capabilities as intelligent agents capable of solving complex problems. However, effective planning in scenarios involving dependencies between API or tool calls-particularly in multi-turn conversations-remains a significant challenge. To address this, we introduce T1, a tool-augmented, multi-domain, multi-turn conversational dataset specifically designed to capture and manage inter-tool dependencies across diverse domains. T1 enables rigorous evaluation of agents' ability to coordinate tool use across nine distinct domains (4 single domain and 5 multi-domain) with the help of an integrated caching mechanism for both short- and long-term memory, while supporting dynamic replanning-such as deciding whether to recompute or reuse cached results. Beyond facilitating research on tool use and planning, T1 also serves as a benchmark for evaluating the performance of open-weight and proprietary large language models. We present results powered by T1-Agent highlighting their ability to plan and reason in complex, tool-dependent scenarios.

Existing molecular machine learning force fields (MLFFs) generally focus on the learning of atoms, molecules, and simple quantum chemical properties (such as energy and force), but ignore the importance of electron density (ED) $\rho(r)$ in accurately understanding molecular force fields (MFFs). ED describes the probability of finding electrons at specific locations around atoms or molecules, which uniquely determines all ground state properties (such as energy, molecular structure, etc.) of interactive multi-particle systems according to the Hohenberg-Kohn theorem. However, the calculation of ED relies on the time-consuming first-principles density functional theory (DFT), which leads to the lack of large-scale ED data and limits its application in MLFFs. In this paper, we introduce EDBench, a large-scale, high-quality dataset of ED designed to advance learning-based research at the electronic scale. Built upon the PCQM4Mv2, EDBench provides accurate ED data, covering 3.3 million molecules. To comprehensively evaluate the ability of models to understand and utilize electronic information, we design a suite of ED-centric benchmark tasks spanning prediction, retrieval, and generation. Our evaluation of several state-of-the-art methods demonstrates that learning from EDBench is not only feasible but also achieves high accuracy. Moreover, we show that learning-based methods can efficiently calculate ED with comparable precision while significantly reducing the computational cost relative to traditional DFT calculations. All data and benchmarks from EDBench will be freely available, laying a robust foundation for ED-driven drug discovery and materials science.

A common problem in LLM evaluation is how to choose a subset of metrics from a full suite of possible metrics. Subset selection is usually done for efficiency or interpretability reasons, and the goal is often to select a "representative" subset of metrics. However, "representative" is rarely clearly defined. In this work, we use ideas from social choice theory to formalize two notions of representation for the selection of a subset of evaluation metrics. We first introduce *positional representation*, which guarantees every alternative is sufficiently represented at every position cutoff. We then introduce *positional proportionality*, which guarantees no alternative is proportionally over- or under-represented by more than a small error at any position. We prove upper and lower bounds on the smallest number of metrics needed to guarantee either of these properties in the worst case. We also study a generalized form of each property that allows for additional input on groups of metrics that must be represented. Finally, we tie theory to practice through real-world case studies on both LLM evaluation and hospital quality evaluation.

Humans rely on multisensory integration to perceive spatial environments, where auditory cues enable sound source localization in three-dimensional space. Despite the critical role of spatial audio in immersive technologies such as VR/AR, most existing multimodal datasets provide only monaural audio, which limits the development of spatial audio generation and understanding. To address these challenges, we introduce MRSAudio, a large-scale multimodal spatial audio dataset designed to advance research in spatial audio understanding and generation. MRSAudio spans four distinct components: MRSLife, MRSSpeech, MRSMusic, and MRSSing, covering diverse real-world scenarios. The dataset includes synchronized binaural and ambisonic audio, exocentric and egocentric video, motion trajectories, and fine-grained annotations such as transcripts, phoneme boundaries, lyrics, scores, and prompts. To demonstrate the utility and versatility of MRSAudio, we establish five foundational tasks: audio spatialization, and spatial text to speech, spatial singing voice synthesis, spatial music generation and sound event localization and detection. Results show that MRSAudio enables high-quality spatial modeling and supports a broad range of spatial audio research. Demos and dataset access are available at https://mrsaudio.github.io.

Academic poster generation is a crucial yet challenging task in scientific communication, requiring the compression of long-context interleaved documents into a single, visually coherent page. To address this challenge, we introduce Paper2Poster, the first benchmark and metric suite for poster generation, which pairs recent conference papers with author-designed posters and evaluates outputs on (i) Visual Quality—semantic alignment with human posters, (ii) Textual Coherence—language fluency, (iii) Holistic Assessment—six fine-grained aesthetic and informational criteria scored by a VLM-as-judge, and notably (iv) PaperQuiz—the poster’s ability to convey core paper content as measured by VLMs answering generated quizzes. Building on this benchmark, we propose PosterAgent, a top‐down, visual‐in‐the‐loop multi‐agent pipeline: the (a) Parser distills the paper into a structured asset library; the (b) Planner aligns text–visual pairs into a binary‐tree layout that preserves reading order and spatial balance; and the (c) Painter–Commenter loop refines each panel by executing rendering code and using VLM feedback to eliminate overflow and ensure alignment. In our comprehensive evaluation, we find that GPT‐4o outputs—though visually appealing at first glance—often exhibit noisy text and poor PaperQuiz scores; We find that reader engagement is the primary aesthetic bottleneck, as human‐designed posters rely largely on visual semantics to convey meaning. Our fully open‐source Paper2Poster pipeline outperforms GPT‐4o–based systems across nearly all metrics while consuming 87 \% fewer tokens. These findings chart clear directions for the next generation of fully automated poster‐generation models.

Visual grounding in 3D is the key for embodied agents to localize language-referred objects in open-world environments. However, existing benchmarks are limited to indoor focus, single-platform constraints, and small scale. We introduce 3EED, a multi-platform, multi-modal 3D grounding benchmark featuring RGB and LiDAR data from vehicle, drone, and quadruped platforms. We provide over 128,000 objects and 22,000 validated referring expressions across diverse outdoor scenes -- 10x larger than existing datasets. We develop a scalable annotation pipeline combining vision-language model prompting with human verification to ensure high-quality spatial grounding. To support cross-platform learning, we propose platform-aware normalization and cross-modal alignment techniques, and establish benchmark protocols for in-domain and cross-platform evaluations. Our findings reveal significant performance gaps, highlighting the challenges and opportunities of generalizable 3D grounding. The 3EED dataset and benchmark toolkit are released to advance future research in language-driven 3D embodied perception.

This paper investigates the occlusion problems in virtual try-on (VTON) tasks. According to how they affect the try-on results, the occlusion issues of existing VTON methods can be grouped into two categories: (1) Inherent Occlusions, which are the ghosts of the clothing from reference input images that exist in the try-on results. (2) Acquired Occlusions, where the spatial structures of the generated human body parts are disrupted and appear unreasonable. To this end, we analyze the causes of these two types of occlusions, and propose a novel mask-free VTON framework based on our analysis to deal with these occlusions effectively. In this framework, we develop two simple-yet-powerful operations: (1) The background pre-replacement operation prevents the model from confusing the target clothing information with the human body or image background, thereby mitigating inherent occlusions. (2) The covering-and-eliminating operation enhances the model's ability of understanding and modeling human semantic structures, leading to more realistic human body generation and thus reducing acquired occlusions. Moreover, our method is highly generalizable, which can be applied in in-the-wild scenarios, and our proposed operations can also be easily integrated into different generative network architectures (e.g., GANs and diffusion models) in a plug-and-play manner. Extensive experiments on three VTON datasets validate the effectiveness and generalization ability of our method. Both qualitative and quantitative results demonstrate that our method outperforms recently proposed VTON benchmarks.

Out-of-distribution (OOD) detection and segmentation are crucial for deploying machine learning models in safety-critical applications such as autonomous driving and robot-assisted surgery. While prior research has primarily focused on unimodal image data, real-world applications are inherently multimodal, requiring the integration of multiple modalities for improved OOD detection. A key challenge is the lack of supervision signals from unknown data, leading to overconfident predictions on OOD samples. To address this challenge, we propose Feature Mixing, an extremely simple and fast method for synthesizing multimodal outliers with theoretical support, which can be further optimized to help the model better distinguish between in-distribution (ID) and OOD data. Feature Mixing is modality-agnostic and applicable to various modality combinations. Additionally, we introduce CARLA-OOD, a new multimodal dataset for OOD segmentation, featuring synthetic OOD objects across diverse scenes and weather conditions. Extensive experiments on SemanticKITTI, nuScenes, CARLA-OOD datasets, and the MultiOOD benchmark demonstrate that Feature Mixing achieves state-of-the-art performance with a $10 \times$ to $370 \times$ speedup. Our source code and dataset are available at https://github.com/mona4399/FeatureMixing.

Deep Neural Networks (DNNs) have demonstrated remarkable performance across various domains, including computer vision and natural language processing. However, they often struggle to accurately quantify their predictions' uncertainty, limiting their broader adoption in critical industrial applications. Uncertainty Quantification (UQ) for Deep Learning seeks to address this challenge by providing methodologies to improve the reliability of uncertainty estimates. While numerous techniques have been proposed, a unified tool remains lacking that offers a seamless workflow for evaluating and integrating these methods. To bridge this gap, we introduce **Torch-Uncertainty**, a *PyTorch* and *Lightning* framework designed to streamline the training and evaluation of DNNs with UQ techniques. In this paper, we outline the foundational principles of our library and present comprehensive experimental results that benchmark a diverse set of UQ methods across classification, segmentation, and regression tasks. Our library is available at: https://github.com/ENSTA-U2IS-AI/torch-uncertainty.

Existing code generation benchmarks primarily evaluate functional correctness, with limited attention to code efficiency, and they are often restricted to a single language such as Python. To address this gap, we introduce EffiBench‑X, the first large‑scale multi‑language benchmark specifically designed for robust efficiency evaluation of LLM‑generated code. EffiBench‑X supports Python, C++, Java, JavaScript, Ruby, and Go, and comprises competitive programming tasks paired with human‑expert solutions as efficiency baselines. Evaluating state‑of‑the‑art LLMs on EffiBench‑X reveals that while models frequently generate functionally correct code, they consistently underperform human experts in efficiency. Even the most efficient LLM‑generated solutions (e.g., Qwen3‑32B) achieve only around 62% of human efficiency on average, with significant language‑specific variation: models tend to perform better in Python, Ruby, and JavaScript than in Java, C++, and Go (e.g., DeepSeek‑R1’s Python code is markedly more efficient than its Java code). These findings highlight the need for research into optimization‑oriented methods to improve the efficiency of LLM‑generated code across diverse languages. The dataset and evaluation infrastructure are publicly available at https://github.com/EffiBench/EffiBench-X.git and https://huggingface.co/datasets/EffiBench/effibench-x.

Large Language Models (LLMs) are transforming search engines into Conversational Search Engines (CSE). Consequently, Search Engine Optimization (SEO) is being shifted into Conversational Search Engine Optimization (C-SEO). We are beginning to see dedicated C-SEO methods for modifying web documents to increase their visibility in CSE responses. However, they are often tested only for a limited breadth of application domains; we do not know whether certain C-SEO methods would be effective for a broad range of domains. Moreover, existing evaluations consider only a single-actor scenario where only one web document adopts a C-SEO method; in reality, multiple players are likely to competitively adopt the cutting-edge C-SEO techniques, drawing an analogy from the dynamics we have seen in SEO. We present C-SEO Bench, the first benchmark designed to evaluate C-SEO methods across multiple tasks, domains, and number of actors. We consider two search tasks, question answering and product recommendation, with three domains each. We also formalize a new evaluation protocol with varying adoption rates among involved actors. Our experiments reveal that most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking, which is opposite to what is expected. Instead, traditional SEO strategies, those aiming to improve the ranking of the source in the LLM context, are significantly more effective. We also observe that as we increase the number of C-SEO adopters, the overall gains decrease, depicting a congested and zero-sum nature of the problem.

Instruction tuning is essential for Large Language Models (LLMs) to effectively follow user instructions. To improve training efficiency and reduce data redundancy, recent works use LLM-based scoring functions, e.g., Instruction-Following Difficulty (IFD), to select high–quality instruction-tuning data with scores above a threshold. While these data selection methods often lead to models that can match or even exceed the performance of models trained on the full datasets, we identify two key limitations: (i) they assess quality at the sample level, ignoring token-level informativeness; and (ii) they overlook the robustness of the scoring method, often selecting a sample due to superficial lexical features instead of its true quality. In this work, we propose Token-Selective HIeRarchical Data Selection for Instruction Tuning (T-SHIRT), a novel data selection framework that introduces a new scoring method to include only informative tokens in quality evaluation and also promote robust and reliable samples whose neighbors also show high quality with less local inconsistencies. We demonstrate that models instruction-tuned on a curated dataset (only 5% of the original size) using T-SHIRT can outperform those trained on the entire large-scale dataset by up to 5.48 points on average across eight benchmarks. Across various LLMs and training set scales, our method consistently surpasses existing state-of-the-art data selection techniques, while also remaining both cost-effective and highly efficient. For instance, by using GPT-2 for score computation, we are able to process a dataset of 52k samples in 40 minutes on a single GPU. Our code is available at https://github.com/Dynamite321/T-SHIRT.