NeurIPS 2025 Papers

Vision-Language Foundation Models (VLMs), trained on large-scale multimodal datasets, have driven significant advances in Artificial Intelligence (AI) by enabling rich cross-modal reasoning. Despite their success in general domains, applying these models to medical imaging remains challenging due to the limited availability of diverse imaging modalities and multilingual clinical data. Most existing medical VLMs are trained on a subset of imaging modalities and focus primarily on high-resource languages, thus limiting their generalizability and clinical utility. To address these limitations, we introduce a novel Vietnamese-language multimodal medical dataset consisting of 2,757 whole-body PET/CT volumes from independent patients and their corresponding full-length clinical reports. This dataset is designed to fill two pressing gaps in medical AI development: (1) the lack of PET/CT imaging data in existing VLMs training corpora, which hinders the development of models capable of handling functional imaging tasks; and (2) the underrepresentation of low-resource languages, particularly the Vietnamese language, in medical vision-language research. To the best of our knowledge, this is the first dataset to provide comprehensive PET/CT-report pairs in Vietnamese. We further introduce a training framework to enhance VLMs' learning, including data augmentation and expert-validated test sets. We conduct comprehensive experiments benchmarking state-of-the-art VLMs on downstream tasks, including medical report generation and visual question answering. The experimental results show that incorporating our dataset significantly improves the performance of existing VLMs. However, despite these advancements, the models still underperform on clinically critical criteria, particularly the diagnosis of lung cancer, indicating substantial room for future improvement. We believe this dataset and benchmark will serve as a pivotal step in advancing the development of more robust VLMs for medical imaging, particularly in low-resource languages, and improving their clinical relevance in Vietnamese healthcare.

The issue-resolving task, where a model generates patches to fix real-world bugs, has emerged as a key benchmark for evaluating the capabilities of large language models (LLMs). While SWE-bench has become the dominant benchmark in this domain, it suffers from several limitations: it has not been updated since its release, is restricted to only 12 repositories, and relies heavily on manual effort for constructing test instances and setting up executable environments, significantly limiting its scalability. We present SWE-bench-Live, a live-updatable benchmark designed to address these limitations. SWE-bench-Live currently includes 1,890 tasks derived from real GitHub issues created since 2024, spanning 223 repositories. Each task is accompanied by a dedicated Docker image to ensure reproducible execution. Additionally, we introduce an automated curation pipeline that streamlines the entire process from instance creation to environment setup, removing manual bottlenecks and enabling scalability and continuous updates. We evaluate a range of state-of-the-art models and agent frameworks on SWE-bench-Live, offering detailed empirical insights into their real-world bug-fixing capabilities. By providing a fresh, diverse, and executable benchmark grounded in live repository activity, SWE-bench-Live supports reliable, large-scale assessment of code LLMs and code agents in realistic development settings.

Offline Learning from Observations (LfO) focuses on enabling agents to imitate expert behavior using datasets that contain only expert state trajectories and separate transition data with suboptimal actions. This setting is both practical and critical in real-world scenarios where direct environment interaction or access to expert action labels is costly, risky, or infeasible. Most existing LfO methods attempt to solve this problem through state or state-action occupancy matching. They typically rely on pretraining a discriminator to differentiate between expert and non-expert states, which could introduce errors and instability—especially when the discriminator is poorly trained. While recent discriminator-free methods have emerged, they generally require substantially more data, limiting their practicality in low-data regimes. In this paper, we propose IOSTOM ($\textit{Imitation from Observation via State Transition Occupancy Matching}$), a novel offline LfO algorithm designed to overcome these limitations. Our approach formulates a learning objective based on the joint state visitation distribution. A key distinction of IOSTOM is that it first excludes actions entirely from the training objective. Instead, we learn an $\textit{implicit policy}$ that models transition probabilities between states, resulting in a more compact and stable optimization problem. To recover the expert policy, we introduce an efficient action inference mechanism that $\textit{avoids training an inverse dynamics model}$. Extensive empirical evaluations across diverse offline LfO benchmarks show that IOSTOM substantially outperforms state-of-the-art methods, demonstrating both improved performance and data efficiency.

Vision-Language Models (VLMs) acquire real-world knowledge and general reasoning ability through Internet-scale image-text corpora. They can augment robotic systems with scene understanding and task planning, and assist visuomotor policies that are trained on robot trajectory data. We explore the reverse paradigm — using rich, real, multi-modal robot trajectory data to enhance and evaluate VLMs. In this paper, we present Robo2VLM, a Visual Question Answering (VQA) dataset generation framework for VLMs. Given a human tele-operated robot demonstration with video and robot data, Robo2VLM derives ground-truth from non-visual and non-descriptive sensory modalities, such as end-effector pose, gripper aperture, and force sensing. Based on these modalities, it segments the robot trajectory into a sequence of manipulation phases. At each phase, Robo2VLM uses scene and interaction understanding to identify 3D properties of the robot, task goal, and the target object. The properties are used to generate representative VQA queries – images with textural multiple-choice questions – based on spatial, goal-conditioned, and interaction reasoning question templates. We use a subset of Open X-Embodiment to generate Robo2VLM-1, a large-scale in-the-wild dataset with 684,710 questions based on 463 distinct scenes and 3,396 robotic manipulation tasks from 176k real robot trajectories. Results suggest that Robo2VLM-1 can benchmark and improve VLM capabilities in spatial and interaction reasoning.

Object compositing offers significant promise for augmented reality (AR) and embodied intelligence applications. Existing approaches predominantly focus on single-image scenarios or intrinsic decomposition techniques, facing challenges with multi-view consistency, complex scenes, and diverse lighting conditions. Recent inverse rendering advancements, such as 3D Gaussian and diffusion-based methods, have enhanced consistency but are limited by scalability, heavy data requirements, or prolonged reconstruction time per scene. To broaden its applicability, we introduce MV-CoLight, a two-stage framework for illumination-consistent object compositing in both 2D images and 3D scenes. Our novel feed-forward architecture models lighting and shadows directly, avoiding the iterative biases of diffusion-based methods. We employ a Hilbert curve-based mapping to align 2D image inputs with 3D Gaussian scene representations seamlessly. To facilitate training and evaluation, we further introduce a large-scale 3D compositing dataset. Experiments demonstrate state-of-the-art harmonized results across standard benchmarks and our dataset, as well as casually captured real-world scenes demonstrate the framework's robustness and wide generalization.

Advancing code reasoning in large language models (LLMs) is fundamentally limited by the scarcity of high-difficulty datasets, especially those with verifiable input-output test cases necessary for rigorous solution validation at scale. We introduce rStar-Coder, which significantly improves LLM code reasoning capabilities by constructing a large-scale, verified dataset of 418K competition-level code problems, 580K long-reasoning solutions along with rich test cases of varying difficulty. This is achieved through three core contributions: (1) we curate competitive programming code problems and solutions to synthesize new, solvable problems; (2) we introduce a reliable input-output test case synthesis pipeline that decouples the generation into a three-step input generation method and a mutual verification mechanism for effective output labeling; (3) we augment problems with high-quality, test-case-verified long-reasoning solutions. Extensive experiments on Qwen models (1.5B-14B) across various code reasoning benchmarks demonstrate the superiority of rStar-Coder dataset, achieving leading performance comparable to frontier reasoning LLMs with significantly smaller model sizes. On LiveCodeBench, rStar-Coder improves Qwen2.5-7B from 17.4% to an impressive 57.3%, and Qwen2.5-14B from 23.3% to 62.5%, surpassing o3-mini (low) by 3.1%. On the more challenging USA Computing Olympiad, our 7B model achieves an average pass@1 accuracy of 16.15%, outperforming the frontier-level QWQ-32B. rStar-Coder dataset is publicly available at https://huggingface.co/datasets/microsoft/rStar-Coder.

What does it mean for two videos to be similar? Videos may appear similar when judged by the actions they depict, yet entirely different if evaluated based on the locations where they were filmed. While humans naturally compare videos by taking different aspects into account, this ability has not been thoroughly studied and presents a challenge for models that often depend on broad global similarity scores. Large Multimodal Models (LMMs) with video understanding capabilities open new opportunities for leveraging natural language in comparative video tasks. We introduce Concept-based Video Similarity estimation (ConViS), a novel task that compares pairs of videos by computing interpretable similarity scores across a predefined set of key semantic concepts. ConViS allows for human-like reasoning about video similarity and enables new applications such as concept-conditioned video retrieval. To support this task, we also introduce ConViS-Bench, a new benchmark comprising carefully annotated video pairs spanning multiple domains. Each pair comes with concept-level similarity scores and textual descriptions of both differences and similarities. Additionally, we benchmark several state-of-the-art models on ConViS, providing insights into their alignment with human judgments. Our results reveal significant performance differences on ConViS, indicating that some concepts present greater challenges for estimating video similarity. We believe that ConViS-Bench will serve as a valuable resource for advancing research in language-driven video understanding.

Transformers have become the de facto architecture for a wide range of machine learning tasks, particularly in large language models (LLMs). Despite their remarkable performance, many challenges remain in training deep transformer networks, especially regarding the position of the layer normalization. While Pre-Norm structures facilitate more stable training owing to their stronger identity path, they often lead to suboptimal performance compared to Post-Norm. In this paper, we propose **HybridNorm**, a simple yet effective hybrid normalization strategy that integrates the advantages of both Pre-Norm and Post-Norm. Specifically, HybridNorm employs QKV normalization within the attention mechanism and Post-Norm in the feed-forward network (FFN) of each transformer block. We provide both theoretical insights and empirical evidence to demonstrate that HybridNorm improves the gradient flow and the model robustness. Extensive experiments on large-scale transformer models, including both dense and sparse variants, show that HybridNorm consistently outperforms both Pre-Norm and Post-Norm approaches across multiple benchmarks. These findings highlight the potential of HybridNorm as a more stable and effective technique for improving the training and performance of deep transformer models. Code is available at https://github.com/BryceZhuo/HybridNorm.

Quantum computing is an emerging field recognized for the significant speedup it offers over classical computing through quantum algorithms. However, designing and implementing quantum algorithms pose challenges due to the complex nature of quantum mechanics and the necessity for precise control over quantum states. Despite the significant advancements in AI, there has been a lack of datasets specifically tailored for this purpose. In this work, we introduce QCircuitBench, the first benchmark dataset designed to evaluate AI's capability in designing and implementing quantum algorithms in the form of quantum circuit codes. Unlike using AI for writing traditional codes, this task is fundamentally more complicated due to highly flexible design space. Our key contributions include: 1. A general framework which formulates the key features of quantum algorithm design task for Large Language Models. 2. Implementation for quantum algorithms from basic primitives to advanced applications, spanning 3 task suites, 25 algorithms, and 120,290 data points. 3. Automatic validation and verification functions, allowing for iterative and interactive evaluation without human inspection. 4. Promising potential as a training dataset through primitive fine-tuning results. We observed several interesting experimental phenomena: fine-tuning does not always outperform few-shot learning, and LLMs tend to exhibit consistent error patterns. QCircuitBench provides a comprehensive benchmark for AI-driven quantum algorithm design, while also revealing some limitations of LLMs in this domain.

The global rise in wildfire frequency and intensity over the past decade underscores the need for improved fire monitoring techniques. To advance deep learning research on wildfire detection and its associated human health impacts, we introduce **SmokeViz**, a large-scale machine learning dataset of smoke plumes in satellite imagery. The dataset is derived from expert annotations created by smoke analysts at the National Oceanic and Atmospheric Administration, which provide coarse temporal and spatial approximations of smoke presence. To enhance annotation precision, we propose **pseudo-label dimension reduction (PLDR)**, a generalizable method that applies pseudo-labeling to refine datasets with mismatching temporal and/or spatial resolutions. Unlike typical pseudo-labeling applications that aim to increase the number of labeled samples, PLDR maintains the original labels but increases the dataset quality by solving for intermediary pseudo-labels (IPLs) that align each annotation to the most representative input data. For SmokeViz, a parent model produces IPLs to identify the single satellite image within each annotations time window that best corresponds with the smoke plume. This refinement process produces a succinct and relevant deep learning dataset consisting of over 160,000 manual annotations. The SmokeViz dataset is expected to be a valuable resource to develop further wildfire-related machine learning models and is publicly available at \url{https://noaa-gsl-experimental-pds.s3.amazonaws.com/index.html#SmokeViz/}.

Occlusion is one of the fundamental challenges in crowd counting. In the community, various data-driven approaches have been developed to address this issue, yet their effectiveness is limited. This is mainly because most existing crowd counting datasets on which the methods are trained are based on passive cameras, restricting their ability to fully sense the environment. Recently, embodied navigation methods have shown significant potential in precise object detection in interactive scenes. These methods incorporate active camera settings, holding promise in addressing the fundamental issues in crowd counting. However, most existing methods are designed for indoor navigation, showing unknown performance in analyzing complex object distribution in large-scale scenes, such as crowds. Besides, most existing embodied navigation datasets are indoor scenes with limited scale and object quantity, preventing them from being introduced into dense crowd analysis. Based on this, a novel task, Embodied Crowd Counting (ECC), is proposed to count the number of persons in a large-scale scene actively. We then build up an interactive simulator, the Embodied Crowd Counting Dataset (ECCD), which enables large-scale scenes and large object quantities. A prior probability distribution approximating a realistic crowd distribution is introduced to generate crowds. Then, a zero-shot navigation method (ZECC) is proposed as a baseline. This method contains an MLLM-driven coarse-to-fine navigation mechanism, enabling active Z-axis exploration, and a normal-line-based crowd distribution analysis method for fine counting. Experimental results show that the proposed method achieves the best trade-off between counting accuracy and navigation cost. Code can be found at https://github.com/longrunling/ECC?.

Stellarators are magnetic confinement devices under active development to deliver steady-state carbon-free fusion energy. Their design involves a high-dimensional, constrained optimization problem that requires expensive physics simulations and significant domain expertise. Recent advances in plasma physics and open-source tools have made stellarator optimization more accessible. However, broader community progress is currently bottlenecked by the lack of standardized optimization problems with strong baselines and datasets that enable data-driven approaches, particularly for quasi-isodynamic (QI) stellarator configurations, considered as a promising path to commercial fusion due to their inherent resilience to current-driven disruptions. Here, we release an open dataset of diverse QI-like stellarator plasma boundary shapes, paired with their ideal magnetohydrodynamic (MHD) equilibria and performance metrics. We generated this dataset by sampling a variety of QI fields and optimizing corresponding stellarator plasma boundaries. We introduce three optimization benchmarks of increasing complexity: (1) a single-objective geometric optimization problem, (2) a "simple-to-build" QI stellarator, and (3) a multi-objective ideal-MHD stable QI stellarator that investigates trade-offs between compactness and coil simplicity. For every benchmark, we provide reference code, evaluation scripts, and strong baselines based on classical optimization techniques. Finally, we show how learned models trained on our dataset can efficiently generate novel, feasible configurations without querying expensive physics oracles. By openly releasing the dataset (https://huggingface.co/datasets/proxima-fusion/constellaration) along with benchmark problems and baselines (https://github.com/proximafusion/constellaration), we aim to lower the entry barrier for optimization and machine learning researchers to engage in stellarator design and to accelerate cross-disciplinary progress toward bringing fusion energy to the grid.

Data valuation techniques quantify each training example's contribution to model performance, providing a principled basis for data cleaning, acquisition, and selection. Existing valuation methods remain inadequate: \emph{model-based} techniques depend on a single fitted model and inherit its biases, while \emph{algorithm-based} approaches like Data Shapley scale poorly due to their need to train multiple models. Recent work has proposed model-agnostic alternatives based on Wasserstein distance between the training set and a clean reference set, but exact computation is expensive and approximations often misrank examples. We introduce KAIROS, a model-agnostic framework that values examples by their contribution to the Maximum Mean Discrepancy (MMD) between the training set and a clean reference distribution. Unlike Wasserstein methods, MMD admits a closed-form solution that requires no approximations and is scalable to large datasets. Additionally, KAIROS enables efficient online valuation: adding a new batch of $m$ examples requires only $O(mN)$ computation to update all scores, compared to $O(N^2)$ in prior work where $N$ is the training set size. Empirical evaluations on noise, mislabeling, and poisoning benchmarks show that KAIROS consistently outperforms state-of-the-art baselines in both accuracy and runtime. On ImageNet, KAIROS achieves up to 15 $\times$ speedup over the fastest baseline while maintaining superior data valuation quality. Our results demonstrate that model-agnostic methods can match or exceed model-based approaches in performance while scaling to large datasets.

Naive DDIM inversion process usually suffers from a trajectory deviation issue, i.e., the latent trajectory during reconstruction deviates from the one during inversion. To alleviate this issue, previous methods either learn to mitigate the deviation or design cumbersome compensation strategy to reduce the mismatch error, exhibiting substantial time and computation cost. In this work, we present a nearly free-lunch method (named FreeInv) to address the issue more effectively and efficiently. In FreeInv, we randomly transform the latent representation and keep the transformation the same between the corresponding inversion and reconstruction time-step. It is motivated from a statistical perspective that an ensemble of DDIM inversion processes for multiple trajectories yields a smaller trajectory mismatch error on expectation. Moreover, through theoretical analysis and empirical study, we show that FreeInv performs an efficient ensemble of multiple trajectories. FreeInv can be freely integrated into existing inversion-based image and video editing techniques. Especially for inverting video sequences, it brings more significant fidelity and efficiency improvements. Comprehensive quantitative and qualitative evaluation on PIE benchmark and DAVIS dataset shows that FreeInv remarkably outperforms conventional DDIM inversion, and is competitive among previous state-of-the-art inversion methods, with superior computation efficiency.

Mental visualization, the ability to construct and manipulate visual representations internally, is a core component of human cognition and plays a vital role in tasks involving reasoning, prediction, and abstraction. Despite the rapid progress of Multimodal Large Language Models (MLLMs), current benchmarks primarily assess passive visual perception, offering limited insight into the more active capability of internally constructing visual patterns to support problem solving. Yet mental visualization is a critical cognitive skill in humans, supporting abilities such as spatial navigation, predicting physical trajectories, and solving complex visual problems through imaginative simulation. To bridge this gap, we introduce Hyperphantasia, a synthetic benchmark designed to evaluate the mental visualization abilities of MLLMs through four carefully constructed puzzles. Each task is procedurally generated and presented at three difficulty levels, enabling controlled analysis of model performance across increasing complexity. Our comprehensive evaluation of state-of-the-art models reveals a substantial gap between the performance of humans and MLLMs. Additionally, we explore the potential of reinforcement learning to improve visual simulation capabilities. Our findings suggest that while some models exhibit partial competence in recognizing visual patterns, robust mental visualization remains an open challenge for current MLLMs.

Code-switching (CS) is the alternating use of two or more languages within a conversation or utterance, often influenced by social context and speaker identity. This linguistic phenomenon poses challenges for Automatic Speech Recognition (ASR) systems, which are typically designed for a single language and struggle to handle multilingual inputs. The growing global demand for multilingual applications, including Code-Switching ASR (CSASR), Text-to-Speech (TTS), and Cross-Lingual Information Retrieval (CLIR), highlights the inadequacy of existing monolingual datasets. Although some code-switching datasets exist, most are limited to bilingual mixing within homogeneous ethnic groups, leaving a critical need for a large-scale, diverse benchmark akin to ImageNet in computer vision. To bridge this gap, we introduce \textbf{LinguaMaster}, a multi-agent collaboration framework specifically designed for efficient and scalable multilingual data synthesis. Leveraging this framework, we curate \textbf{SwitchLingua}, the first large-scale multilingual and multi-ethnic code-switching dataset, including: (1) 420K CS textual samples across 12 languages, and (2) over 80 hours of audio recordings from 174 speakers representing 18 countries/regions and 63 racial/ethnic backgrounds, based on the textual data. This dataset captures rich linguistic and cultural diversity, offering a foundational resource for advancing multilingual and multicultural research. Furthermore, to address the issue that existing ASR evaluation metrics lack sensitivity to code-switching scenarios, we propose the \textbf{Semantic-Aware Error Rate (SAER)}, a novel evaluation metric that incorporates semantic information, providing a more accurate and context-aware assessment of system performance. Benchmark experiments on SwitchLingua with state-of-the-art ASR models reveal substantial performance gaps, underscoring the dataset’s utility as a rigorous benchmark for CS capability evaluation. In addition, SwitchLingua aims to encourage further research to promote cultural inclusivity and linguistic diversity in speech technology, fostering equitable progress in the ASR field. LinguaMaster (Code): github.com/Shelton1013/SwitchLingua, SwitchLingua (Data): https://huggingface.co/datasets/Shelton1013/SwitchLingua_text, https://huggingface.co/datasets/Shelton1013/SwitchLingua_audio

The rapid advancement of generative models, such as GANs and Diffusion models, has enabled the creation of highly realistic synthetic images, raising serious concerns about misinformation, deepfakes, and copyright infringement. Although numerous Artificial Intelligence Generated Image (AIGI) detectors have been proposed, often reporting high accuracy, their effectiveness in real-world scenarios remains questionable. To bridge this gap, we introduce AIGIBench, a comprehensive benchmark designed to rigorously evaluate the robustness and generalization capabilities of state-of-the-art AIGI detectors. AIGIBench simulates real-world challenges through four core tasks: multi-source generalization, robustness to image degradation, sensitivity to data augmentation, and impact of test-time pre-processing. It includes 23 diverse fake image subsets that span both advanced and widely adopted image generation techniques, along with real-world samples collected from social media and AI art platforms. Extensive experiments on 11 advanced detectors demonstrate that, despite their high reported accuracy in controlled settings, these detectors suffer significant performance drops on real-world data, limited benefits from common augmentations, and nuanced effects of pre-processing, highlighting the need for more robust detection strategies. By providing a unified and realistic evaluation framework, AIGIBench offers valuable insights to guide future research toward dependable and generalizable AIGI detection.

Graph Neural Networks (GNNs) are models that leverage the graph structure to transmit information between nodes, typically through the message-passing operation. While widely successful, this approach is well-known to suffer from representational collapse as the number of layers increases and insensitivity to the information contained at distant and poorly connected nodes. In this paper, we present a unified view of on the appearance of these issues through the lens of vanishing gradients, using ideas from linear control theory for our analysis. We propose an interpretation of GNNs as recurrent models and empirically demonstrate that a simple state-space formulation of an GNN effectively alleviates these issues at no extra trainable parameter cost. Further, we show theoretically and empirically that (i) Traditional GNNs are by design prone to extreme gradient vanishing even after few layers; (ii) Feature collapse is directly related to the mechanism causing vanishing gradients; (iii) Long-range modeling is most easily achieved by a combination of graph rewiring and vanishing gradient mitigation. We believe our work will help bridge the gap between the recurrent and graph neural network literature and will unlock the design of new deep and performant GNNs.

Score-based generative models (SGMs) have emerged as one of the most popular classes of generative models. A substantial body of work now exists on the analysis of SGMs, focusing either on discretization aspects or on their statistical performance. In the latter case, bounds have been derived, under various metrics, between the true data distribution and the distribution induced by the SGM, often demonstrating polynomial convergence rates with respect to the number of training samples. However, these approaches adopt a largely approximation theory viewpoint, which tends to be overly pessimistic and relatively coarse. In particular, they fail to fully explain the empirical success of SGMs or capture the role of the optimization algorithm used in practice to train the score network. To support this observation, we first present simple experiments illustrating the concrete impact of optimization hyperparameters on the generalization ability of the generated distribution. Then, this paper aims to bridge this theoretical gap by providing the first algorithmic- and data-dependent generalization analysis for SGMs. In particular, we establish bounds that explicitly account for the optimization dynamics of the learning algorithm, offering new insights into the generalization behavior of SGMs. Our theoretical findings are supported by empirical results on several datasets.