NeurIPS 2026 论文列表

Spreadsheets are among the most widely used tools for computations, and formulas are their core building blocks. Besides agentic spreadsheet tools, proactive formula suggestions enable significant productivity gains and are now available in popular spreadsheet software (Microsoft Excel and Google Sheets). Such suggestions pose interesting challenges: the right context must be collected and represented for a machine learning model, which must then be able to write a correct formula, in a way that is fast and cheap enough to trigger on every $\texttt{=}$ (or subsequent keystrokes for auto-completion) without taking users out of their editing flow. Surprisingly, there is no public benchmark for formula suggestions on real spreadsheets. In this paper, we therefore (1) introduce fxbench as a manually curated benchmark of 503 formula suggestion tasks based on Sheetpedia, (2) break down the formula suggestion problem into four capabilities---gather context, understand context, understand intent, write formula---that a formula suggestion system requires, and (3) evaluate existing and new approaches to understand bottlenecks in these capabilities that pose interesting research directions. Specifically, we annotate each benchmark with the cells that describe intent to determine the theoretical performance limit of a given context selection strategy, we evaluate performance of small and large models on different representations of context---including a dense format with high theoretical coverage at low token counts---and we fine-tune smaller language models to show that we can learn to understand such dense formats. Additionally, we perform quantitative and qualitative analysis on failure modes to understand bottlenecks in the formula writing capability---the best configuration only achieved 58.8\% correct suggestions.

Commercial video generation systems such as Seedance2.0 and Veo3.1 have rapidly improved, strengthening the view that video generators may be evolving into "world simulators." Yet the community still lacks a benchmark that directly tests whether a model can reason about how an observed world should evolve over time. We introduce WorldReasonBench, which reframes video generation evaluation as world-state prediction: given an initial state and an action, can a model generate a future video whose state evolution remains physically, socially, logically, and informationally consistent? WorldReasonBench contains 436 curated test cases with structured ground-truth QA annotations spanning four reasoning dimensions and 22 subcategories. We evaluate generated videos with a human-aligned two-part methodology: Process-aware Reasoning Verification uses structured QA and reasoning-phase diagnostics to detect temporal and causal failures, while Multi-dimensional Quality Assessment scores reasoning quality, temporal consistency, and visual aesthetics for ranking and reward modeling. We further introduce WorldRewardBench, a preference benchmark with approximately 6K expert-annotated pairs over 1.4K videos, supporting pair-wise and point-wise reward-model evaluation. Across modern video generators, our results expose a persistent gap between visual plausibility and world reasoning: videos can look convincing while failing dynamics, causality, or information preservation. We will release our benchmarks and evaluation toolkit to support community research on genuinely world-aware video generation.

Video world models aim to generate temporally coherent visual observations conditioned on actions, but often suffer from 3D geometric inconsistencies such as spatial drift, object deformation, and perspective violations, which limit their reliability in downstream tasks. We propose WorldPrism, a post-training framework that improves 3D consistency via reinforcement learning without modifying the base model architecture. The central challenge lies in designing a reliable reward signal. To address this, we introduce Bidirectional Geometry-Flow Verification, a bidirectional cross-space mechanism that evaluates consistency through two complementary paths: (i) lifting 2D optical flow into 3D to measure position-space agreement, and (ii) projecting reconstructed 3D geometry to match the point correspondence. By leveraging these two independent and complementary signals, this design mitigates the blind spots of unilateral 2D evaluation. We further incorporate hierarchical temporal evaluation to capture both local and long-range geometric consistency. Experiments on diverse datasets demonstrate that WorldPrism achieves state-of-the-art 3D consistency while preserving visual quality and generative diversity.

A solid understanding of the predictive success of deep neural networks (DNNs) remains elusive. For example, while many data-dependent metrics exist that seek to predict model quality of DNNs, these metrics may be expensive to compute for large datasets and/or may be impossible to compute when training datasets are not released. A practical and effective predictor of model quality involves analyzing DNN weight matrices, as it is known that a heavy-tailed spectral distribution often correlates with strong model quality. As such metrics lack a rigorous statistical derivation, in this work we aim to understand why they perform well, by employing a probabilistic framework to derive the marginal likelihood for the trained weights of a DNN. Across large-scale convolutional DNNs and LLMs, we find that this marginal likelihood predicts model quality. To understand the role heavy-tailed spectra play in predictive performance, we derive the limiting log-marginal likelihood for spectra that follow the heavy-tailed High-Temperature Marchenko-Pastur (HTMP) distribution. We show that this limiting quantity is convex in a *heaviness* parameter, and we derive an optimal *heaviness* that empirically predicts over-fitting for large-scale DNNs during training.

Video Large Language Models (Video-LLMs) have made rapid progress on temporal video understanding, yet many fail at a basic perceptual primitive: signed image-plane motion direction. On simple videos of a single object moving left, right, up, or down, most Video-LLMs perform near chance, with above-chance cases largely attributable to prediction biases rather than genuine direction understanding. We call this failure directional motion blindness. We localize the failure by tracing motion direction information through the Video-LLM pipeline. Motion direction remains linearly accessible from the vision encoder, projector, and LLM hidden states, but the readout fails to bind this signal to the correct verbal answer option, revealing a direction binding gap. Although synthetic motion direction instruction tuning reduces this gap on the source domain, motion direction concept vector analysis shows that visual complexity weakens the signal magnitude and limits out-of-domain generalization. We introduce MODIRECT, a dataset family for motion direction instruction tuning and evaluation, and DeltaDirect, a diagnosis-driven, projector-level objective that predicts normalized 2-D motion vectors from adjacent-frame feature deltas. On MODIRECT-SYNBENCH, instruction tuning with DeltaDirect improves motion direction accuracy from 25.9% to 85.4%. On MODIRECT-REALBENCH, DeltaDirect improves real-world motion direction accuracy by 21.9 points over the vanilla baseline without real-world tuning data, while preserving standard video-understanding performance.

Parameter-efficient fine-tuning (PEFT) and probing enable adaptation of foundation models using only a small number of trainable parameters, making it attractive for video understanding where annotation and computation are expensive. However, video PEFT has focused on adapting image-pretrained models, while standard PEFT methods can also be applied to video representations. These settings are rarely compared and both confine temporal reasoning to a single component of the model, leaving open how temporal context should be distributed across backbone, PEFT and probe. In this work we provide a systematic study of model adaptation strategies for video understanding. We evaluate methods across appearance-focused, motion-focused and spatially dense settings, with a particular focus on scenarios with limited data where parameter-efficiency is most beneficial. Our results provide new insights into PEFT and probing across settings and demonstrate the importance of temporal context allocation for effective video adaptation.

Safety-aligned large language models often lose their safety behaviors after fine-tuning, even when safety data are included. While this fragility is well documented, its underlying cause remains unclear. We propose that formulaic refusal strings function as *outlier modes* in the model's output distribution, i.e., low-support behaviors that are easy to fit but structurally unstable under subsequent fine-tuning, and that this is an important source of safety fragility. Controlled experiments verify this: fixed refusals are acquired and forgotten much like arbitrary constant strings, in contrast to prompt-grounded natural responses. Motivated by this insight, we improve safety retention by constructing natural, context-aware safety supervision that keeps safety responses within the model's existing output distribution and grounded in semantics. We instantiate this principle in both supervised and preference-based alignment settings, using perplexity under the original model as a practical proxy for distributional naturalness. Across multiple models and fine-tuning regimes, our approach improves both immediate safety and its retention under continued fine-tuning without sacrificing helpfulness, suggesting that naturalness is a useful principle for durable safety alignment.

Medical vision--language models (VLMs) are attracting growing interest as a foundation for clinical image interpretation, retrieval, and decision support. However, current evaluations largely emphasize aggregate accuracy, retrieval, or classification performance, offering limited insight into whether these models possess the fine-grained semantic understanding required for safe and reliable use in high-stakes clinical settings. In this work, we introduce \textbf{MedTEC-Bench}, a diagnostic benchmark for probing semantic understanding in medical VLMs across five levels of increasing specificity, from broad modality recognition to fine-grained findings, and negation. MedTEC-Bench spans five medical imaging modalities: chest radiography, brain MRI, retinal fundus photography, dermoscopy, and histopathology. It combines controlled semantic probes with a suite of metrics designed to reveal failures hidden by aggregate benchmarks. Evaluating a diverse set of VLMs, from broad biomedical models to clinical-specialist models, we find that existing systems often fail on basic yet clinically meaningful semantic distinctions, despite strong reported performance on standard tasks. Our results suggest that current medical VLM evaluations may overestimate vision-grounded clinical understanding and highlight the need for targeted semantic benchmarks before such models are deployed in high-stakes medical workflows.

Practitioners using machine learning on noisy tabular data can often find simpler, more interpretable models with the same accuracy as black box models. In some cases, this is because the noise in the data induces a large Rashomon set, that is, a large set of near-optimal models, which then frequently contains simpler models. However, our understanding of this effect from discrete hypothesis classes --- the Rashomon set's size, the diversity of its predictions, and implicit regularization --- leads to a contradiction in continuous hypothesis spaces, and thus does not directly establish whether noise should necessarily induce simpler or more diverse models. We reconcile the seemingly conflicting effects by showing geometrically that noise drives simplicity through separate mechanisms for linear, logistic, and generalized additive models. We introduce the Rashomon reducibility cost to quantify whether a given model simplification is possible within the Rashomon set, and characterize how this cost changes under general data properties such as label noise. For practitioners working with noisy tabular data, our results suggest that simpler, more interpretable models will be included in the Rashomon set, so that black box models will generally not be needed.

Streaming video understanding models must answer queries at any moment during an ongoing stream, using only what they have observed so far and under fixed memory and computation budgets. Existing methods address this by adding memory banks, retrieval modules, or visual token compression to preserve long-range history. However, strong recent-window baselines show that indiscriminate history injection can dilute current-scene perception, suggesting that the key challenge is not whether to use memory, but how to allocate it selectively. We formulate this as budgeted online latent evidence allocation and propose **SelectStream**, a selective latent-memory framework that keeps the current observation directly visible to a frozen VLM while exposing historical information only through a compact, query-conditioned evidence budget. Three coordinated mechanisms govern when to write, what to preserve, and how to retrieve: surprise-driven adaptive windowing, priority-preserving consolidation, and query-conditioned graph reasoning over a fixed-capacity latent memory graph. Retrieved evidence is calibrated and injected as latent tokens for answer generation, without replaying frames or growing the context with stream length. Experimental results show that SelectStream achieves strong online streaming performance and preserves general video understanding, reaching 82.67\% on StreamingBench, 67.03\% on OVO-Bench, and 74.4\% average accuracy on offline video benchmarks, while outperforming strong recent-window baselines and prior streaming memory methods.

Long-horizon video generation is increasingly moving toward autoregressive rollout, where each newly generated segment becomes part of the evidence used for future prediction. Recent causal decoders, long-context training, and KV-cache mechanisms have extended feasible duration, but most inference pipelines still decide the visible past implicitly through recent windows, compressed caches, or external memory. This raises a direct question: **can the choice of which past frames or segments to show to the decoder be treated as an explicit test-time decision?** We answer this question with \textsc{ReCR}, a training-free method that formulates autoregressive video decoding as representation-guided context rewriting. At each step, \textsc{ReCR} builds a selected visible context from the generated history by using internal decoder representations to favor context units that match the global rollout state, avoid redundancy with the recent continuation, preserve temporal boundary coverage, and provide useful intermediate evidence. The selected units are then mapped onto a compact causal axis before decoding the next unit. Across long-horizon text-to-video benchmarks, \textsc{ReCR} consistently improves multiple autoregressive backbones under matched visible-token budgets. Empirical analyses such as representation-space diagnostics and transition-count prompt switching further show that improving what remains visible is an effective path toward more stable long-video decoding. Our code is available at https://anonymous.4open.science/r/anonymous-ReCR-97EB/

Modern visual backbones, despite their architectural diversity, largely reduce relational reasoning to pairwise message passing on graphs and therefore primarily capture 2-order interactions. However, visual understanding often depends on latent yet crucial higher-order semantic correlations among multiple regions, which cannot be adequately expressed by pairwise modeling alone. We first formalize this limitation by proving an irreducibility result: on a fixed vertex set, there exist nonlinear hypergraph message passing layers that cannot be represented exactly by any graph message passing layer unless the hypergraph satisfies highly restrictive structural conditions. This shows that higher-order correlation modeling offers a fundamental expressive advantage rather than a simple approximation to graph-based backbones. Therefore, we propose Vision Correlator (ViC), a general-purpose visual backbone built on multi-order hypergraphs. ViC consists of correlation induction and correlation propagation. In the correlation induction stage, we propose an Anchor-based Hypergraph Generation strategy that uses anchors as semantic centers and incorporates topological information to generate scalable hypergraph structures. In the correlation propagation stage, we further design Differential Mixed Aggregation to achieve efficient hyperedge representation and vertex feature updating, and introduce hyperedge dropout to improve the stability of structure learning. Extensive experiments with isotropic and pyramid variants show that our ViC achieves a better accuracy-efficiency trade-off than strong Transformer and graph-based baselines. Specifically, compared to the Vision Transformer baseline, ViC achieves up to 76\% parameter reduction, 93\% FLOPs reduction, and delivers up to a 5.1\% increase in accuracy on ImageNet.

Multi-task scene understanding models typically require modifications to the vision encoder or complex decoders for cross-task interaction, which limits their flexibility and compatibility with off-the-shelf vision foundation models (VFMs). This paper introduces Virtual Task Prompting (VTP), which frames multi-task prompting as retrieving task-adaptive subsets from a shared virtual state, where both the state and accessing process itself are optimizable in an end-to-end manner. Concretely, VTP maintains a set of virtual tasks alongside the model, and at each layer, learnable routing matrices read an actual task composition as input for interaction with image tokens, then write the processed results back, forming explicit learning pathways that capture the intricacies among tasks before and after the computation of each encoder block. Since VTP keeps the backbone clean, it integrates seamlessly with recent VFMs like DINOv3, advancing the state of the art on Pascal-Context and NYUD-v2. Stress test on Taskonomy with 13 heterogeneous tasks further shows VTP's robustness in extreme settings. Ablations and analysis confirm that VTP's gains stem mainly from its prompting design rather than merely increased capacity.

Long-form video understanding (LVU) requires models to reason over hour-long narratives while preserving fine-grained spatiotemporal details across extended temporal horizons. Although recent multimodal large language models (MLLMs) have achieved strong performance on short-video tasks, their reasoning capabilities degrade substantially as video duration and information density increase. We present VideoSleuth, a narrative-centric agentic framework for LVU built on video- and audio-native MLLMs. VideoSleuth organizes its memory around the four narrative elements-time, place, person, and event-and equips the agent with tools explicitly designed to construct and query these elements within a ReAct-style reasoning loop. Unlike prior LVU agents that rely on dense offline preprocessing or image-only VLMs, VideoSleuth performs perception on demand using a video-audio native perceptual backbone, jointly leveraging visual and auditory cues to support dialogue-based identity grounding, long-range event localization, and fine-grained evidence retrieval. On four standard LVU benchmarks, VideoSleuth consistently outperforms prior on-demand grounding agents and achieves competitive accuracy with the strongest dense-preprocessing baseline, while consuming at least $5.2\times$ fewer end-to-end tokens than DVD-style pipelines spend on offline preprocessing alone.

Recent advances in Multi-modal Large Language Models (MLLMs) have significantly improved video understanding. However, real-world scenarios often require going beyond isolated video inputs, giving rise to the emerging paradigm of video deep research, which involves open-world reasoning through integration of external knowledge. Despite this progress, existing approaches are limited by their inability to perform effective inter-video interaction and their reliance on static, one-shot datasets that yield noisy and suboptimal supervision. To address these challenges, we propose VideoSailor, an open-world video deep research agent powered by a trajectory-to-policy flywheel that iteratively converts generated reasoning trajectories into high-quality off-policy guidance to mitigate the limitations of low-quality on-policy rollouts. For systematic evaluations, we further introduce Omni-BrowseComp, a benchmark that emphasizes omni-modal evidence aggregation, cross-video reasoning, and fine-grained temporal grounding with reasoning-relevant segment annotations. Extensive experiments demonstrate that VideoSailor significantly improves performance on complex multi-hop and cross-video reasoning tasks, establishing a scalable and effective paradigm for open-world video deep research.

Multimodal queries can require different types of reasoning. Some can be answered via *perceptual* reasoning, extracting information directly from the visual signal, while others require *compositional* reasoning that combines observations or *deliberative* reasoning that evaluates competing hypotheses. However, many existing methods apply a uniform reasoning strategy across queries, leading to unnecessary computation on simple tasks and insufficient reasoning on complex ones. We introduce **Video-FLAIR**, a training framework that learns to select the appropriate reasoning mode for each query using reinforcement learning. During training, the model generates responses under all three modes for the same prompt, enabling direct comparison. A composite reward compares these responses to favor the most effective one based on correctness, grounding, and cost, while discouraging unsupported or misaligned deliberation. This yields a supervision signal for learning adaptive reasoning without per-query annotations. Video-FLAIR improves accuracy over the Qwen2.5-VL base model by **+5.4** on MathVista, **+4.8** on Video-Holmes, and **+4.8** on Video-MMMU, while reducing average token usage to **95** compared to **417** for always-thinking baselines.

We address geometric video alignment for the same scene captured by multiple freely moving cameras. Existing methods, which are based on image stitching or video stabilization, are typically limited to two cameras or must assume the restrictive setting of a fixed linear ordering of the cameras. We propose MViS, a lightweight test-time optimization framework for joint multi-view alignment with temporal consistency. MViS uses a spatial-transformer-inspired net to jointly process frames from all views within a short temporal window and predict per-view, per-time homographies. Optimization proceeds in two stages: geometry-guided alignment followed by appearance-based refinement, with both stages enforcing multi-view and temporal consistency across the sequence. The method requires no scene-specific hyperparameter tuning, handles arbitrary and time-varying view-intersection graphs, and remains robust under rapid camera motion. On a standard two-view video benchmark, MViS matches or surpasses the state of the art. To facilitate evaluation in more complex scenarios, we introduce an annotated dataset with more than two freely moving cameras and establish the first benchmark for this setting. Our code and dataset will be released upon acceptance.

Large Vision-Language Models (LVLMs) show strong video understanding, yet hallucinations persist, with outputs that misalign with the video content. This stems from two primary causes. First, consecutive frames are often semantically similar, so the model can miss visual details. Second, LVLMs may rely on language priors and pay less attention to visual evidence. To mitigate video hallucinations, we propose **VidHalluDoctor**, a training framework that enhances fine-grained discrimination and reduces reliance on language bias. VidHalluDoctor builds video difference samples to improve visual detail recognition, counterfactual samples to reduce prior bias, and consistency-aware reinforcement learning with designed rewards that evaluate both reasoning quality and consistency between the reasoning and answer. To better evaluate video hallucinations, we introduce **VidHalluEva**, a comprehensive benchmark comprising 10,000 videos that covers three progressive hallucination types across diverse scenarios and multi-view settings. Extensive experiments demonstrate that VidHalluDoctor significantly reduces hallucinations while maintaining strong video understanding. Code and dataset will be released.

Video scene text editing aims to replace text appearing on scene surfaces in a video, such as storefront signs, whiteboards, and product labels, while preserving the surrounding content, motion, and camera dynamics. Although scene text editing has been extensively studied for still images, its video counterpart remains underdeveloped. Existing resources do not provide paired edits over real-world videos, current protocols rarely measure whether the edited region reads as the target string over time, and there is no large-scale open-source benchmark for systematic comparison. To close these gaps, we introduce ViTeX-Bench, a comprehensive benchmark suite for high-fidelity and temporally consistent video scene text editing. Specifically, we first built ViTeX-Dataset, containing 387 real-world 720p videos with precise text-region masks and editing instructions; the 230-training split provides the first such resource with paired editing results built through a semi-automatic pipeline, with the rest forming the evaluation set for standardized benchmarking. In ViTeX-Bench, we propose a three-axis evaluation protocol: text correctness, visual quality, and edit locality, which contains a total of 13 metrics that combine character-level OCR signals with motion and preservation metrics. Our benchmarking experiments indicate that all eight leading video editing models, including both public and commercial models, fail in distinct ways: they either flicker or drift over time, or simply fail to produce the requested editing effects. To further validate the efficacy of our proposed dataset, we trained ViTeX-Edit-14B on the paired training split with a motion-aligned glyph-video conditioning stream. We show that, trained on only 230 high-quality paired videos, ViTeX-Edit-14B achieves the strongest CharAcc among video-native editors (0.688, +11.1\% over VideoPainter) and leads three ranked temporal metrics, reducing the previous best by 2.1\% (Flicker$_f$), 6.6\% (Flicker$_c$), and 1.9\% (Warp$_c$), respectively.