NeurIPS 2025 Papers

Efficient deployment of small language models (SLMs) is essential for numerous real-world applications with stringent latency constraints.While previous work on SLM design has primarily focused on reducing the number of parameters to achieve parameter-optimal SLMs, parameter efficiency does not necessarily translate into proportional real-device speed-ups. This work aims to identify the key determinants of SLMs' real-device latency and offer generalizable principles and methodologies for SLM design and training when real-device latency is the primary consideration. Specifically, we identify two central architectural factors: depth–width ratios and operator choices. The former is crucial for small-batch-size latency, while the latter affects both latency and large-batch-size throughput. In light of this, we first study latency-optimal depth–width ratios, with the key finding that although deep–thin models generally achieve better accuracy under the same parameter budget, they may not lie on the accuracy–latency trade-off frontier. Next, we explore emerging efficient attention alternatives to evaluate their potential as candidate building operators. Using the identified promising operators, we construct an evolutionary search framework to automatically discover latency-optimal combinations of these operators within hybrid SLMs, thereby advancing the accuracy–latency frontier. In addition to architectural improvements, we further enhance SLM training using a weight normalization technique that enables more effective weight updates and improves final convergence. This technique can serve as a generalizable component for future SLMs. Combining these methods, we introduce a new family of hybrid SLMs, called Nemotron-Flash, which significantly advances the accuracy–efficiency frontier of state-of-the-art SLMs, e.g., achieving over +5.5\% average accuracy, 1.3$\times$/1.9$\times$ lower latency, and 18.7$\times$/45.6$\times$ higher throughput compared to Qwen3-1.7B/0.6B, respectively.

Linear recurrent neural networks (RNNs) and state-space models (SSMs) such as Mamba have become promising alternatives to softmax-attention as sequence mixing layers in Transformer architectures. Current models, however, do not exhibit the full state-tracking expressivity of RNNs because they rely on channel-wise (i.e. diagonal) sequence mixing. In this paper, we investigate parameterizations of a large class of dense linear RNNs as fixed-points of parallelizable diagonal linear RNNs. The resulting models can naturally trade expressivity for efficiency at a fixed number of parameters and achieve state-of-the-art results on the state-tracking benchmarks $A_5$ and $S_5$, while matching performance on copying and other tasks.

Recent advances in Multimodal Large Language Models (MLLMs) have enabled the development of mobile agents that can understand visual inputs and follow user instructions, unlocking new possibilities for automating complex tasks on mobile devices. However, applying these models to real-world mobile scenarios remains a significant challenge due to the long-horizon task execution, difficulty in error recovery, and the cold-start problem in unfamiliar environments. To address these challenges, we propose MobileUse, a GUI agent designed for robust and adaptive mobile task execution. To improve resilience in long-horizon tasks and dynamic environments, we introduce a hierarchical reflection architecture that enables the agent to self-monitor, detect, and recover from errors across multiple temporal scales—ranging from individual actions to overall task completion—while maintaining efficiency through a Reflection-on-Demand strategy. To tackle cold-start issues, we further introduce a proactive exploration module, which enriches the agent’s understanding of the environment through self-planned exploration. Evaluations on the AndroidWorld and AndroidLab benchmarks demonstrate that MobileUse establishes new state-of-the-art performance, achieving success rates of 62.9% and 44.2%, respectively. To facilitate real-world applications, we release an out-of-the-box toolkit for automated task execution on physical mobile devices, which is available at https://github.com/MadeAgents/mobile-use.

This paper investigates the integration of response time data into human preference learning frameworks for more effective reward model elicitation. While binary preference data has become fundamental in fine-tuning foundation models, generative AI systems, and other large-scale models, the valuable temporal information inherent in user decision-making remains largely unexploited. We propose novel methodologies to incorporate response time information alongside binary choice data, leveraging the Evidence Accumulation Drift Diffusion (EZ) model, under which response time is informative of the preference strength. We develop Neyman-orthogonal loss functions that achieve oracle convergence rates for reward model learning, matching the theoretical optimal rates that would be attained if the expected response times for each query were known a priori. Our theoretical analysis demonstrates that for linear reward functions, conventional preference learning suffers from error rates that scale exponentially with reward magnitude. In contrast, our response time-augmented approach reduces this to polynomial scaling, representing a significant improvement in sample efficiency. We extend these guarantees to non-parametric reward function spaces, establishing convergence properties for more complex, realistic reward models. Our extensive experiments validate our theoretical findings in the context of preference learning over images.

Recent advances in scene understanding have leveraged multimodal large language models (MLLMs) for 3D reasoning by capitalizing on their strong 2D pretraining. However, the lack of explicit 3D data during MLLM pretraining limits 3D representation capability. In this paper, we investigate the 3D-awareness of MLLMs by evaluating multi-view correspondence and reveal a strong positive correlation between the quality of 3D-aware representation and downstream task performance. Motivated by this, we propose 3DRS, a framework that enhances MLLM 3D representation learning by introducing supervision from pretrained 3D foundation models. Our approach aligns MLLM visual features with rich 3D knowledge distilled from 3D models, effectively improving scene understanding. Extensive experiments across multiple benchmarks and MLLMs—including visual grounding, captioning, and question answering—demonstrate consistent performance gains. Code will be released to facilitate future research.

Physics-informed neural networks (PINN) have achieved notable success in solving partial differential equations (PDE), yet solving the Navier-Stokes equations (NSE) with complex boundary conditions remains a challenging task. In this paper, we introduce a novel Hybrid Boundary PINN (HB-PINN) method that combines a pretrained network for efficient initialization with a boundary-constrained mechanism. The HB-PINN method features a primary network focused on inner domain points and a distance metric network that enhances predictions at the boundaries, ensuring accurate solutions for both boundary and interior regions. Comprehensive experiments have been conducted on the NSE under complex boundary conditions, including the 2D cylinder wake flow and the 2D blocked cavity flow with a segmented inlet. The proposed method achieves state-of-the-art (SOTA) performance on these benchmark scenarios, demonstrating significantly improved accuracy over existing PINN-based approaches.

Pseudo-label-based Semi-Supervised Learning (SSL) often suffers from classifier bias, particularly under class imbalance, as inaccurate pseudo-labels tend to exacerbate existing biases towards majority classes. Existing methods, such as \textit{CDMAD}\cite{cdmad}, utilize simplistic reference inputs—typically uniform or blank-colored images—to estimate and correct this bias. However, such simplistic references fundamentally ignore realistic statistical information inherent to real datasets, specifically typical color distributions, texture details, and frequency characteristics. This lack of \emph{statistical representativeness} can lead the model to inaccurately estimate its inherent bias, limiting the effectiveness of bias correction, particularly under severe class imbalance or substantial distribution mismatches between labeled and unlabeled datasets. To overcome these limitations, we introduce the \textbf{FARAD} (Fourier-Adapted Reference for Accurate Debiasing) System. This system utilizes random-phase images, constructed by preserving the amplitude spectrum of real data while randomizing the phase spectrum. This strategy ensures two critical properties: (1) \textbf{Semantic Irrelevance}, as randomizing phase removes any structural or recognizable semantic cues, and (2) \textbf{Statistical Representativeness}, as preserving the amplitude spectrum maintains realistic textures, color distributions, and frequency characteristics. Grounded theoretically in classical Fourier analysis, the FARAD System provides a robust, accurate estimation of per-class biases. Furthermore, computational efficiency is enhanced through optimized real-to-complex (R2C) batched Fast Fourier Transforms (FFTs). Comprehensive experiments demonstrate that our approach, significantly improves minority-class accuracy and overall SSL performance, particularly under challenging imbalance scenarios, compared with existing reference-based bias correction methods.

Shapley values have emerged as a critical tool for explaining which features impact the decisions made by machine learning models. However, computing exact Shapley values is difficult, generally requiring an exponential (in the feature dimension) number of model evaluations. To address this, many model-agnostic randomized estimators have been developed, the most influential and widely used being the KernelSHAP method (Lundberg & Lee, 2017). While related estimators such as unbiased KernelSHAP (Covert & Lee, 2021) and LeverageSHAP (Musco & Witter, 2025) are known to satisfy theoretical guarantees, bounds for KernelSHAP have remained elusive. We describe a broad and unified framework that encompasses KernelSHAP and related estimators constructed using both with and without replacement sampling strategies. We then prove strong non-asymptotic theoretical guarantees that apply to all estimators from our framework. This provides, to the best of our knowledge, the first theoretical guarantees for KernelSHAP and sheds further light on tradeoffs between existing estimators. Through comprehensive benchmarking on small and medium dimensional datasets for Decision-Tree models, we validate our approach against exact Shapley values, consistently achieving low mean squared error with modest sample sizes. Furthermore, we make specific implementation improvements to enable scalability of our methods to high-dimensional datasets. Our methods, tested on datasets such MNIST and CIFAR10, provide consistently better results compared to the KernelSHAP library.

Facial personalization faces challenges to maintain identity fidelity without disrupting the foundation model's prompt consistency. The mainstream personalization models employ identity embedding to integrate identity information within the attention mechanisms. However, our preliminary findings reveal that identity embeddings compromise the effectiveness of other tokens in the prompt, thereby limiting high prompt consistency and attribute-level controllability. Moreover, by deactivating identity embedding, personalization models still demonstrate the underlying foundation models' ability to control facial attributes precisely. It suggests that such foundation models' knowledge can be leveraged to cure the ill-aligned prompt consistency of personalization models. Building upon these insights, we propose FreeCure, a framework that improves the prompt consistency of personalization models with their latent foundation models' knowledge. First, by setting a dual inference paradigm with/without identity embedding, we identify attributes (e.g., hair, accessories, etc.) for enhancements. Second, we introduce a novel foundation-aware self-attention module, coupled with an inversion-based process to bring well-aligned attribute information to the personalization process. Our approach is training-free, and can effectively enhance a wide array of facial attributes; and it can be seamlessly integrated into existing popular personalization models based on both Stable Diffusion and FLUX. FreeCure has consistently shown significant improvements in prompt consistency across these facial personalization models while maintaining the integrity of their original identity fidelity.

Existing medical RAG systems mainly leverage knowledge from medical knowledge bases, neglecting the crucial role of experiential knowledge derived from similar patient cases - a key component of human clinical reasoning. To bridge this gap, we propose DoctorRAG, a RAG framework that emulates doctor-like reasoning by integrating both explicit clinical knowledge and implicit case-based experience. DoctorRAG enhances retrieval precision by first allocating conceptual tags for queries and knowledge sources, together with a hybrid retrieval mechanism from both relevant knowledge and patient. In addition, a Med-TextGrad module using multi-agent textual gradients is integrated to ensure that the final output adheres to the retrieved knowledge and patient query. Comprehensive experiments on multilingual, multitask datasets demonstrate that DoctorRAG significantly outperforms strong baseline RAG models and gains improvements from iterative refinements. Our approach generates more accurate, relevant, and comprehensive responses, taking a step towards more doctor-like medical reasoning systems.

Vision-language models (VLMs) have made great strides in addressing temporal understanding tasks, which involve characterizing visual changes across a sequence of images. However, recent works have suggested that when making predictions, VLMs may rely on static feature biases, such as background or object features, rather than dynamic visual changes. Static feature biases are a type of shortcut and can contribute to systematic prediction errors on downstream tasks; as a result, identifying and characterizing error-inducing static feature biases is critical prior to real-world model deployment. Existing approaches for identifying such systematic failure modes in trained models (i) are typically designed for non-temporal settings and (ii) are challenging to evaluate in temporal settings due to the lack of quantitative evaluation frameworks. In this work, we address these challenges by introducing TRoVe, an automated approach for discovering error-inducing static feature biases learned by temporal VLMs. Given a trained VLM and an annotated validation dataset associated with a downstream classification task, TRoVe extracts candidate static features from the dataset and scores each feature by (i) the effect of the feature on classification errors as well as (ii) the extent to which the VLM relies on the feature when making predictions. In order to quantitatively evaluate TRoVe, we introduce an evaluation framework consisting of 101 trained temporal VLMs paired with ground-truth annotations for learned static feature biases. We use this framework to demonstrate that TRoVe can accurately identify error-inducing static feature biases in VLMs, achieving a 28.6% improvement over the closest baseline. Finally, we apply TRoVe to 7 off-the-shelf VLMs and 2 temporal understanding tasks, surfacing previously-unknown static feature biases and demonstrating that knowledge of learned biases can aid in improving model performance at test time. Our code is available at https://github.com/Stanford-AIMI/TRoVe.

Virtual furniture synthesis, a critical task in image composition, aims to seamlessly integrate reference objects into indoor scenes while preserving geometric coherence and visual realism. Despite its significant potential in home design applications, this field remains underexplored due to two major challenges: the absence of publicly available and ready-to-use benchmarks hinders reproducible research, and existing image composition methods fail to meet the stringent fidelity requirements for realistic furniture placement. To address these issues, we introduce RoomBench, a ready-to-use benchmark dataset for virtual furniture synthesis, comprising 7,298 training pairs and 895 testing samples across 27 furniture categories. Then, we propose RoomEditor, a simple yet effective image composition method that employs a parameter-sharing dual U-Net architecture, ensuring better feature consistency by sharing weights between dual branches. Technical analysis reveals that conventional dual-branch architectures generally suffer from inconsistent intermediate features due to independent processing of reference and background images. In contrast, RoomEditor enforces unified feature learning through shared parameters, thereby facilitating model optimization for robust geometric alignment and maintaining visual consistency. Experiments show our RoomEditor is superior to state-of-the-arts, while generalizing directly to diverse objects synthesis in unseen scenes without task-specific fine-tuning. Our dataset and code are available at https://github.com/stonecutter-21/roomeditor.

Zero-shot Human-Object Interaction detection aims to localize humans and objects in an image and recognize their interaction, even when specific verb-object pairs are unseen during training. Recent works have shown promising results using prompt learning with pretrained vision-language models such as CLIP, which align natural language prompts with visual features in a shared embedding space. However, existing approaches still fail to handle the *visual complexity of interaction*—including (1) *intra-class visual diversity*, where instances of the same verb appear in diverse poses and contexts, and (2) *inter-class visual entanglement*, where distinct verbs yield visually similar patterns. To address these challenges, we propose **VDRP**, a framework for *Visual Diversity and Region-aware Prompt learning*. First, we introduce a visual diversity-aware prompt learning strategy that injects group-wise visual variance into the context embedding. We further apply Gaussian perturbation to encourage the prompt to capture diverse visual variations of a verb. Second, we retrieve region-specific concepts from the human, object, and union regions. These are used to augment the diversity-aware prompt embeddings, yielding region-aware prompts that improve verb-level discrimination. Experiments on the HICO-DET benchmark demonstrate that our method achieves state-of-the-art performance under four zero-shot evaluation settings, effectively addressing both intra-class diversity and inter-class visual entanglement. Code is available at https://github.com/mlvlab/VDRP.

Computer vision for animals holds great promise for wildlife research but often depends on large-scale data, while existing collection methods rely on controlled capture setups. Recent data-driven approaches show the potential of single-view, non-invasive analysis, yet current animal video datasets are limited—offering as few as 2.4K 15-frame clips and lacking key processing for animal-centric 3D/4D tasks. We introduce an automated pipeline that mines YouTube videos and processes them into object-centric clips, along with auxiliary annotations valuable for downstream tasks like pose estimation, tracking, and 3D/4D reconstruction. Using this pipeline, we amass 30K videos (2M frames)—an order of magnitude more than prior works. To demonstrate its utility, we focus on 4D quadruped animal reconstruction task. To support this task, we present Animal4D, a benchmark of 230 manually filtered sequences with 11K frames showcasing clean, diverse animal motions. We evaluate state-of-the-art model-based and model-free methods on Animal4D, finding that 2D metrics favor the former despite unrealistic 3D shapes, while the latter yields more natural reconstructions but scores lower—revealing a gap in current evaluation. To address this, we enhance a recent model-free approach with sequence-level optimization, establishing the first 4D animal reconstruction baseline. Together, our pipeline, benchmark, and baseline aim to advance large-scale, markerless 4D animal reconstruction and related tasks from in-the-wild videos. Code and datasets are available at https://github.com/briannlongzhao/Animal-Video-Processing.

Large language models (LLMs) have transformed natural language processing, but their reliable deployment requires effective uncertainty quantification (UQ). Existing UQ methods are often heuristic and lack a fully probabilistic foundation. This paper begins by providing a theoretical justification for the role of perturbations in UQ for LLMs. We then introduce a dual random walk perspective, modeling input–output pairs as two Markov chains with transition probabilities defined by semantic similarity. Building on this, we propose a fully probabilistic framework based on an inverse model, which quantifies uncertainty by evaluating the diversity of the input space conditioned on a given output through systematic perturbations. Within this framework, we define a new uncertainty measure, Inv-Entropy. A key strength of our framework is its flexibility: it supports various definitions of uncertainty measures, embeddings, perturbation strategies, and similarity metrics. We also propose GAAP, a perturbation algorithm based on genetic algorithms, which enhances the diversity of sampled inputs. In addition, we introduce a new evaluation metric, Temperature Sensitivity of Uncertainty (TSU), which directly assesses uncertainty without relying on correctness as a proxy. Extensive experiments demonstrate that Inv-Entropy outperforms existing semantic UQ methods.

Constrained hybrid-action reinforcement learning (RL) promises to learn a safe policy within a parameterized action space, which is particularly valuable for safety-critical applications involving discrete-continuous hybrid action spaces. However, existing hybrid-action RL algorithms primarily focus on reward maximization, which faces significant challenges for tasks involving both cost constraints and hybrid action spaces. In this work, we propose a novel Constrained Hybrid-action Policy Optimization algorithm (CHPO) to address the problems of constrained hybrid-action RL. Concretely, we rethink the limitations of hybrid-action RL in handling safe tasks with parameterized action spaces and reframe the objective of constrained hybrid-action RL by introducing the concept of Constrained Parameterized-action Markov Decision Process (CPMDP). Subsequently, we present a constrained hybrid-action policy optimization algorithm to confront the constrained hybrid-action problems and conduct theoretical analyses demonstrating that the CHPO converges to the optimal solution while satisfying safety constraints. Finally, extensive experiments demonstrate that the CHPO achieves competitive performance across multiple experimental tasks.

Quantifying differences between probability distributions is fundamental to statistics and machine learning, primarily for comparing statistical uncertainty. In contrast, epistemic uncertainty---due to incomplete knowledge---requires richer representations than those offered by classical probability. Imprecise probability (IP) theory offers such models, capturing ambiguity and partial belief. This has driven growing interest in imprecise probabilistic machine learning (IPML), where inference and decision-making rely on broader uncertainty models---highlighting the need for metrics beyond classical probability. This work introduces the Integral Imprecise Probability Metric (IIPM) framework, a Choquet integral-based generalisation of classical Integral Probability Metric to the setting of capacities---a broad class of IP models encompassing many existing ones, including lower probabilities, probability intervals, belief functions, and more. Theoretically, we establish conditions under which IIPM serves as a valid metric and metrises a form of weak convergence of capacities. Practically, IIPM not only enables comparison across different IP models but also supports the quantification of epistemic uncertainty~(EU) within a single IP model. In particular, by comparing an IP model with its conjugate, IIPM gives rise to a new class of epistemic uncertainty measures---Maximum Mean Imprecision (MMI) ---which satisfy key axiomatic properties proposed in the Uncertainty Quantification literature. We validate MMI through selective classification experiments, demonstrating strong empirical performance against established EU measures, and outperforming them when classical methods struggle to scale to a large number of classes. Our work advances both theory and practice in Imprecise Probabilistic Machine Learning, offering a principled framework for comparing and quantifying epistemic uncertainty under imprecision.

Federated Learning (FL) incurs high communication costs in both uplink and downlink. The literature largely focuses on lossy compression of model updates in deterministic FL. In contrast, stochastic (Bayesian) FL considers distributions over parameters, enabling uncertainty quantification, better generalization, and, crucially, inherent communication-regularized training through a mirror-descent structure. In this paper, we consider both uplink and downlink communication in stochastic FL, and propose a communication framework based on remote source generation. Employing Minimal Random Coding (MRC) for remote generation, we allow the server and the clients to sample from local and global posteriors (sources), respectively, rather than transmitting locally sampled updates. The framework encompasses communication-regularized local optimization and principled compression of model updates, leveraging gradually updated prior distributions as side information. Through extensive simulations, we show that our method achieves $5-32\times$ reduction in total communication cost while preserving accuracy. We further analyze the communication cost, refining existing MRC bounds and enabling precise quantification of uplink and downlink trade-offs. We also extend our method to conventional FL via stochastic quantization and prove a contraction property for the biased MRC compressor to facilitate convergence analysis.

Vision-Language Models (VLMs) have demonstrated great potential in interpreting remote sensing (RS) images through language-guided semantic. However, the effectiveness of these VLMs critically depends on high-quality image-text training data that captures rich semantic relationships between visual content and language descriptions. Unlike natural images, RS lacks large-scale interleaved image-text pairs from web data, making data collection challenging. While current approaches rely primarily on rule-based methods or flagship VLMs for data synthesis, a systematic framework for automated quality assessment of such synthetically generated RS vision-language data is notably absent. To fill this gap, we propose a novel score model trained on large-scale RS vision-language preference data for automated quality assessment. Our empirical results demonstrate that fine-tuning CLIP or advanced VLMs (e.g., Qwen2-VL) with the top 30% of data ranked by our score model achieves superior accuracy compared to both full-data fine-tuning and CLIP-score-based ranking approaches. Furthermore, we demonstrate applications of our scoring model for reinforcement learning (RL) training and best-of-N (BoN) test-time scaling, enabling significant improvements in VLM performance for RS tasks. Our code, model, and dataset are publicly available.