NeurIPS 2026 论文列表

Scientific AI agents increasingly retrieve evidence, reason across sources, and synthesize conclusions used in consequential decisions. Yet, their ability to do so in high-stakes domains such as health remains unclear. We introduce SciConBench, a large-scale live benchmark of 9.11K questions and expert-written conclusions from systematic reviews to evaluate open-domain scientific conclusion synthesis. The benchmark draws on an expert-validated automated evaluation pipeline that decomposes conclusions into atomic facts and measures correctness and comprehensiveness via factual precision and recall. To mitigate data leakage, we further introduce SciConHarness, a clean-room evaluation harness that equips agents with controlled web interaction to ensure valid measurement. Evaluating 8 frontier models and deep research agents, we find that factual quality remains low: under clean-room settings, the best agent achieves only a factual F1 of 0.337. Our clean-room setting consistently reduces performance relative to unconstrained evaluation, suggesting that leakage inflates estimates of models' true synthesis capabilities. Finally, we audit consumer-facing agents (e.g., Google AI Overview, OpenEvidence) and find they frequently generate incomplete and sometimes contradictory conclusions, even when the ground-truth answer is available. Overall, our results show that reliable synthesis of scientific conclusions remains an open challenge, and that clean-room evaluation is essential for assessing open-domain AI agents.

Runtime oversight for LLM agents is commonly framed as scalar risk prediction: estimate failure likelihood, confidence, or uncertainty, then intervene once the score crosses a threshold. We argue that this framing targets the wrong object for control. The relevant question is not how likely the agent is to fail if it continues, but whether an available intervention would improve the outcome. Two trajectory prefixes can have the same risk estimate while requiring different actions, because one remains recoverable and the other does not. We formalize this mismatch as target error and identify intervention advantage, the expected utility gain from intervening rather than continuing, as the decision object for oversight. To measure this mismatch, we introduce prefix branching, a same-prefix counterfactual protocol that executes candidate actions from identical trajectory states. Across four benchmarks, action-conditioned control yields regime-dependent gains over scalar routing. In a calibration decomposition, recalibrating the same scalar score improves prediction metrics but leaves control regret unchanged, showing that calibration alone does not repair target error. A simple prefix-only action-conditioned controller substantially reduces regret in the strongest interactive regime, from $0.506$ to $0.110$ on ALFWorld. Gains shrink when interventions are weak or when scalar routing already preserves intervention-relevant information. These results suggest that LLM-agent oversight should move from calibrated risk scoring toward action-conditioned value estimation.

Transformer-based scientific foundation models are increasingly deployed in high-stakes settings, but current architectures give deterministic outputs and provide limited support for calibrated predictive uncertainty. We propose Stochastic Attention, a sample average lightweight inference-time modification that randomizes attention by replacing softmax weights with normalized multinomial samples controlled by a single concentration parameter, and produces predictive ensembles without retraining. To set this parameter, we introduce a calibration objective that matches the stochastic attention output with the target, yielding an efficient univariate post-hoc tuning problem. We evaluate this mechanism on scientific foundation models for weather and time-series forecasting, as well as several regression tasks. Across benchmarks against uncertainty-aware baselines, we find that Sample Average Stochastic Attention achieves the strongest native calibration and the sharpest prediction intervals at comparable calibration, with adaptation costs nearly three orders of magnitude lower than the next-best baseline.

Generative models can produce individually plausible samples while deviating substantially from a reference set in the distribution of key features. For example, a model pretrained on broad drug-like chemical space may generate molecules whose molecular features differ from those of a therapeutic class of interest, such as known antibiotics. Correcting such distributional _miscalibration_ is challenging: direct finetuning on the reference set can overfit and does not control which features are matched. To fill this gap, we introduce kernel Calibrating Generative Models (kCGM). kCGM minimizes a maximum mean discrepancy (MMD) between generated and reference feature distributions using an unbiased score-function estimator, with KL regularization to remain close to the pretrained model. On a reference set of 174 antibiotics, direct finetuning sacrifices chemical validity for feature-distribution matching, whereas kCGM improves target feature matching while increasing validity. We further demonstrate kCGM in protein and DNA generation tasks, showing it can adapt autoregressive, continuous-space diffusion, and discrete diffusion models using only feature-level supervision.

Estimating curvature information from gradients alone is a fundamental problem in machine learning. The challenge becomes particularly acute in settings where only aggregate mini-batch gradients are available, yet curvature information is still required for applications such as preconditioning in optimisation, sampling, and early stopping. Perhaps surprisingly, we show that a simple target-perturbation scheme suffices to recover such curvature information from gradients alone. Our method injects zero-mean noise into the targets, with variance calibrated to the local output curvature. The covariance of the resulting noisy batch gradients then recovers practical curvature matrices, without requiring per-sample gradients, second-order derivatives, or other internal network quantities. For a broad class of non-linear models, our scheme recovers the Generalised Gauss--Newton (GGN) matrix; for feed-forward ReLU networks, it additionally recovers the diagonal blocks of the population Hessian. We demonstrate that the resulting estimator is simple, lightweight, and easy to integrate into existing training pipelines. Experiments demonstrate accurate curvature recovery, faster optimisation when the estimator is used for preconditioning, and the utility of the estimated GGN for early stopping.

Safe Policy Improvement (SPI) aims to improve a baseline policy offline using fixed data while avoiding performance degradation with high probability. Existing SPI methods work in discrete domains but do not extend to continuous control. We propose Calibrated Safe Policy Improvement (Cal-SPI), a Monte Carlo Tree Search-based continuous SPI method that replaces count-based support with calibrated model trust. Cal-SPI learns a deep ensemble dynamics model and calibrates ensemble disagreement on held-out data through a Wilks-style tolerance construction, turning raw uncertainty into a statistical certificate of one-step model error. This certificate defines a hard gate within model-based tree search. We provide a theoretical analysis showing that calibrated model trust and a baseline-relative switching criterion yield a conditional PAC-style improvement bound for gate-certified root decisions. Experiments across three MuJoCo continuous domains show that Cal-SPI avoids performance degradation in low-data regimes and improves over the baseline once calibrated model trust becomes reliable.

Reliable probability estimates are critical in many machine learning applications, yet modern classifiers are often poorly calibrated. Post-hoc calibration provides a simple and widely used solution, but the large number of proposed methods, combined with small-scale and inconsistent evaluations, makes it difficult to determine which approaches are truly effective in practice. We introduce a large-scale, standardized benchmark for post-hoc calibration, covering nearly 2000 experiments across tabular and computer vision tasks, including binary, multiclass, and large-scale classification settings. Our benchmark aggregates predictions from a diverse set of classical models, modern deep learning architectures, and foundation models, and provides unified, reproducible implementations of dozens of calibration methods within a common evaluation framework. We argue that Post-Hoc Improvement (PHI) in proper scoring rules offers a principled alternative to traditional calibration error estimators for comparing post-hoc methods, capturing both calibration quality and potential degradation to the model's predictive performance. Using this framework, we conduct the most comprehensive empirical study of post-hoc calibration to date. Our results reveal consistent patterns across domains: smooth calibration functions outperform binning-based approaches, dedicated multiclass methods are essential in high-dimensional settings, and generic machine learning models are not competitive without calibration-specific design. To facilitate future research, we release all data, code, and evaluation tools, providing a plug-and-play benchmark for developing and comparing calibration methods.

Although recent training-free acceleration methods have achieved promising speedups for Diffusion Transformers, most rely on historical features, either by directly reusing cached features or by forecasting future features from them. Such a history-to-future design can accumulate errors when the prediction horizon becomes long. To address this issue, we propose FuCa, a Cache-the-Future training-free acceleration framework. Instead of extrapolating future information solely from the past, FuCa caches sparsely computed future features and uses them as online revision signals. FuCa revises the latent-state update and the corresponding velocity, and then recovers skipped timesteps through lightweight interpolation. It requires no extra training or model-specific auxiliary modules, while keeping the number of DiT forward evaluations low. Experiments on FLUX.1-dev, Qwen-Image, Wan2.1-1.3B, and Hunyuan Video show that FuCa achieves up to about 5$\times$ speedup for image generation and more than 4.3$\times$ speedup for video generation, with higher reconstruction quality than existing training-free baselines.

AI agents are vulnerable to prompt injection attacks, where malicious content hijacks agent behavior. Among proposed defenses, architectural isolation provides the strongest guarantees by strictly separating trusted task planning from untrusted environment observations. However, applying this design to Computer Use Agents (CUAs), which automate tasks by viewing screens and executing actions, presents a fundamental challenge. Current agents require continuous observation of UI state to determine each action, which conflicts with the isolation required for security. We resolve this tension by demonstrating that UI workflows, while dynamic, are structurally predictable. Single-shot planning, where a trusted planner emits upfront a complete branching plan covering all anticipated runtime states, provides control flow integrity guarantees against arbitrary instruction injections. We introduce NOVA (Navigating via Observation, Verification, and Action) to make this viable in the combinatorially large UI state space, where the plan can invoke a perception model to resolve runtime values such as UI coordinates. We evaluate our design on OSWorld, and retain up to 57\% of the performance of frontier models while improving performance for smaller open-source models by up to 19\%, demonstrating that rigorous security and utility can coexist in CUAs. Although upfront planning prevents instruction injections, we show that additional measures are needed to defend against Branch Steering attacks, where adversaries deceive the perception model into routing execution down attacker-preferred branches of the plan, such as redirecting the agent to a malicious website.

Recently, video generation models have achieved remarkable progress in global visual fidelity. However, when synthesizing complex dynamic content, they frequently produce **sparse subtle anomalies**. Existing video reward models primarily emphasize overall visual quality and text alignment; yet as generation quality improves, these coarse-grained defects become less frequent, while the sparse structural, temporal, and physical anomalies become an increasingly important bottleneck. Therefore, in this paper, we propose Concentrate and Concentrate (CaC), a coarse-to-fine anomaly reward model based on Vision-Language Models. During inference, it first conducts a global temporal scan to anchor anomalous time windows, then performs fine-grained spatial grounding within the localized interval, and finally derives robust judgments via structured spatiotemporal Chain-of-Thought reasoning. To equip the model with these capabilities, we construct the first large-scale generated video anomaly dataset with per-frame bounding-box annotations, temporal anomaly windows, and fine-grained attribution labels. Building on this dataset, we design a three-stage progressive training paradigm. The model initially learns spatial and temporal anchoring through single- and multi-frame supervised fine-tuning, and then is optimized by a reinforcement learning strategy based on two-turn Group Relative Policy Optimization (GRPO). Beyond conventional accuracy rewards, we introduce Temporal and Spatial IoU rewards to supervise the intermediate localization process, effectively guiding the model toward more grounded and interpretable spatiotemporal reasoning. Extensive experiments demonstrate that CaC can stably ``concentrate'' on subtle anomalies, achieving a 25.7\% accuracy improvement on fine-grained anomaly benchmarks and, when used as a reward signal, CaC reduces generated-video anomalies by 11.7\% while improving overall video quality.

Irregularly sampled multivariate time series arise in many real-world domains such as healthcare, climate science, and large-scale monitoring systems. Characterized by asynchronous observations and heterogeneity across variables, such time series pose significant challenges for forecasting. In this work, we introduce Cross-Variable Temporal Attention (CVTA), a short-term, continuous-time forecasting framework that learns prediction functions from a compact, interpretable Markov state summarizing recent temporal dynamics across variables. CVTA uses Variable-Conditioned Embeddings to project heterogeneous variables into variable-specific latent spaces, and applies temporal attention to capture cross-variable dependencies from this state. Furthermore, we demonstrate that CVTA can be integrated as an auxiliary module into existing state-of-the-art prediction models through a dynamic meta-decision model. Extensive experiments on multiple real-world benchmarks show that CVTA achieves strong short-term predictive accuracy, supports efficient online inference, improves existing prediction models when used as an auxiliary module, and learns attention patterns that align well with domain knowledge.

Treatment assignment problems arise wherever limited budget must be allocated to heterogeneous users, with applications ranging from personalized recommendations to online advertising and healthcare. In such settings, individuals exhibit heterogeneous responses to different treatments, making it essential to learn cost-aware personalized treatments. This paper introduces the Cost per Unit Value Equalization Tree (CUVET) algorithm, a novel treatment assignment approach that partitions the user space. Under a diminishing-returns (power-law) assumption, CUVET solves the within-cohort allocation problem by equalizing the marginal cost per unit value across each user group, yielding a closed-form cost-aware treatment assignment suited for large-scale industrial deployment. We also release CUVET-policy, an 86.7-million-impression public benchmark derived from real-world industrial A/B tests, providing an open-source evaluation framework for decision-focused learning. Across a multi-method benchmark suite, the Bayesian variant of CUVET achieves the largest cost-feasible value uplift on the public MT-LIFT dataset +11.1% under LP, vs. at most +1.3% for all baselines, while on CUVET-policy, where treatment effects are sub-percent, CUVET variants are the most efficient and strongest cost-feasible methods.

Offline reinforcement learning (RL) trains policies from static datasets, but deployed agents may later need to attenuate the influence of specific data for privacy, safety, or fairness. This is challenging in the overlap regime, where the forget set and retained data share state-action support: unlearning cannot be uniform as suppressing values on shared support can inadvertently erase behaviors required for high performance. We propose Conservative Unlearning with soft-gating Regularization (CURE), a selective unlearning framework tailored to overlap. CURE assigns each forget trajectory an overlap-interference score via cosine alignment between its temporal difference (TD) semi-gradient and a retained reference gradient, and uses it to soft-gate i) conservative, state-conditional critic suppression relative to a retained baseline and ii) coupled actor updates, selectively modulating forgetting based on interference, reducing impact on aligned data. We provide a first-order analysis showing that CURE descends a forget surrogate while controlling retained drift via alignment-dependent coupling. Across MuJoCo benchmarks and offline RL backbones, CURE consistently improves the forget-utility trade-off over baselines, achieving low auditor-detectable forget influence with near-original returns at a fraction of retraining cost.

Visual reprogramming (VR) efficiently repurposes pre-trained vision-language models for new image classification tasks by adding trainable patterns to inputs without modifying the backbone. Nevertheless, existing VR methods heavily depend on labeled data and thus suffer substantial performance degradation under limited supervision (i.e., scarce labeled images). We introduce semi-supervised learning (SSL) into VR to exploit abundant unlabeled images and enhance data efficiency. However, directly applying generic SSL techniques often amplifies biases in unlabeled posteriors and destabilizes VR training. To address these challenges, we propose **CURE** (**C**onsistency and **U**nlabeled **R**ecalibration with Confidence-Margin **E**nhancement), a semi-supervised VR framework comprising two key components: *(i)* Recalibreted Gaussian Confidence-Margin Soft Weighting to dynamically adjust sample importance; *(ii)* Dual-attribute Consistency Regularization to ensure consistency across attribute prompts. Extensive experiments show that CURE consistently outperforms both existing VR methods and direct SSL extensions, improving average classification accuracy by **5.06** and **6.67** percentage points across twelve widely-used benchmarks for limited supervision tasks. Our code is available at https://anonymous.4open.science/r/CURE-93E2/.

Cross-Domain Recommendation is essential for enabling cross-sell in multi-service platforms; however, limited user overlap and strict cross-service data-sharing constraints often result in little to no ground-truth supervision for cross-domain learning. To bridge this gap, we propose an LLM-based framework for cross-domain recommendation in the non-overlapping user scenario. First, we propose an agentic pipeline that constructs cross-domain pseudo supervision, which is unavailable in the non-overlapping user scenario, by leveraging a user’s source-domain history to generate a target-domain item recommendation. Using this dataset, we further propose \underline{C}oupled \underline{U}ser-G\underline{R}ouped R\underline{E}inforcement Learning (CURE), which induces coupling between in- and cross-domain gradient updates via user-level joint normalization, thereby calibrating synthetic cross-domain updates against reliable in-domain signals. We provide a theoretical analysis showing that user-level coupling stabilizes policy updates under imperfect pseudo supervision and improves cross-domain generalization. Experiments on Amazon Reviews and MovieLens show consistent gains over baselines in cross-domain recommendation, while also improving in-domain performance. An online A/B test demonstrates a statistically significant 56\% lift in click-through rate over existing ML methods.

Debugging CUDA programs has long been challenging because failures often arise from subtle interactions among hardware behavior, compiler decisions, memory hierarchy, and asynchronous execution. More importantly, with the rapid expansion of GPU usage across scientific computing, machine learning, graphics, and systems workloads, CUDA debugging has become more challenging than ever. Current evaluations of LLM-based CUDA programming largely miss this setting: a model can pass correctness tests with repair by degeneration, simplifying the CUDA code into a safer but slower program that abandons the original optimization structure. We introduce CUDABEAVER, a benchmark for CUDA debugging from real failing workspaces produced during LLM-based CUDA generation. Each task provides the broken candidate, native build/test commands, raw error evidence, and a single editable file. CUDABEAVER evaluates whether a fixer truly repairs the failing CUDA code or merely finds a slower test-passing replacement, reporting results by failure category, debugging trajectory, stagnation mode, and performance preservation. We further propose pass@k(M,C,A), a protocol-conditional CUDA debugging metric by making the fixer M, corpus C, and protocol axes Aexplicit. Using this metric across 213 tasks and seven frontier LLMs, we show that protocol-aware evaluation gives a more faithful view of CUDA debugging ability: when performance-loss tolerance is high, fixers appear much stronger, but even a minor stricter performance requirement can sharply reduce measured success, shifting scores by up to 40 percentage points.

Despite remarkable advances, today's AI systems remain narrow in scope, falling short of the flexible, adaptive, and multisensory intelligence that characterizes human capabilities. This gap has fueled longstanding debates about whether AI might one day achieve human-like generality or even consciousness, and whether theories of consciousness can inspire new architectures for AI. This paper presents an early blueprint for implementing a general AI system, CTM-AI, combining the Conscious Turing Machine (CTM), a formal machine model of consciousness, with today's foundation models. CTM-AI contains an enormous number of powerful processors ranging from specialized experts (e.g., vision-language models and APIs) to unspecialized general-purpose learners poised to develop their own expertise. Crucially, for whatever problem must be dealt with, information from many processors is selected, integrated, and exchanged appropriately to solve the task.CTM-AI achieves state-of-the-art accuracy on MUStARD (72.28) and UR-FUNNY (72.13), outperforming multimodal and multi-agent frameworks. On tool-using and agentic tasks, CTM-AI achieves 10+ points of improvement on StableToolBench and WebArena-Lite. Overall, CTM-AI offers a principled, testable blueprint for general AI inspired by a model of consciousness.

Lesion analysis in CT imaging, the task of identifying and delineating abnormal regions across diverse anatomical sites, plays a fundamental role in clinical diagnosis. Existing CT lesion datasets typically focus on a single disease type within a specific region. This design leads to two main limitations. First, co-existing lesions outside the predefined categories are left unannotated, so models may treat real lesions as background during training. Second, lesion categories within any single region follow a long-tailed distribution, in which rare but clinically important categories are underrepresented. However, collecting enough samples for every rare category within a single region is often impractical. We observe that certain lesion categories are rare in one region but occur more frequently in another, motivating the use of cross-region complementarity for lesion detection. In this work, we curate a multi-region CT lesion dataset, namely CT-Lesion, which covers brain, chest, and abdomen CT images. Our CT-Lesion is sourced from a private cohort of 500 clinical CT scans. Seven radiologists spent approximately 1,600 working hours delineating lesions of any type, rather than restricting annotations to a predefined set of categories. As a result, CT-Lesion provides clinically confirmed voxel-level segmentation masks and 28 fine-grained lesion labels. Moreover, we further group the lesions into 9 coarse categories to alleviate data scarcity issues. To demonstrate the effectiveness and utility of CT-Lesion, we benchmark the representative methods on CT-Lesion for lesion segmentation and detection. Furthermore, pretraining on CT-Lesion improves model performance on external benchmarks, including LiTS, KiTS, and LIDC-IDRI, demonstrating its potential as a transferable resource for CT lesion analysis. Our dataset is available at Kaggle.

Indoor context deciphered by ambient sensing underpins many applications, from assisted living to smart buildings, yet remains weakly connected to the language-aligned paradigm of modern multimodal systems. Through billions of IoT devices, ambient WiFi Channel State Information (CSI) captures the indoor context of human activity and environmental dynamics passively via ubiquitous, always-on infrastructure. However, unlike vision or audio, CSI lacks naturally paired text like captions or transcripts, limiting its integration with language-driven AI systems. We present CSI-TextBench, a large-scale dataset and benchmark for aligning CSI data with natural language, unifying 3 sensing benchmarks into 285,790 samples across 7 tasks and pairing them with 2,291 hierarchically structured, label-derived descriptions via many-to-many assignment to construct 2.0 million CSI–text training tuples. Benchmarking 5 alignment methods from 4 paradigm families, we show that contrastive CSI--text alignment matches supervised classifiers without class labels, adapts to unseen classes with few samples, and responds to open-vocabulary natural-language prompts. A single unified model supports compositional scene understanding, answering joint queries over activity, proximity, and identity, achieving 86.9\% top-1 accuracy on 140 composite scenes, outperforming task-specific classifier ensembles. CSI-TextBench enables ubiquitous ambient sensing as a language-indexed perception layer, enabling indoor contextual awareness in language-driven AI systems.