NeurIPS 2026 Papers

Evaluating enterprise LLMs on file-grounded Question Answering (QA) requires large, labeled corpora of realistic enterprise PDFs, yet real documents are sensitive, scarce, and impossible to annotate at scale. Existing synthetic approaches produce visually crude renders, annotate ground truth post-hoc (introducing hallucination risk), or are non-reproducible one-off scripts with no diversity control. We present SynthDocQA, an Intermediate Representation(IR)-driven pipeline that co-generates photorealistic multi-page enterprise PDFs and provenance-grounded evaluation pairs in a single pass. A compact Document Generation IR encodes topic, content schedule, and perturbation profile before rendering begins. Templates are retrieved per page via Contrastive Language-Image Pretraining (CLIP) embedding-based cross-modal search over a 10,000-layout library; a 15-stage physical degradation pipeline models five scan quality tiers from pristine to poor. Every QA assertion is derived analytically from the same structured artifact that populated the page, making hallucination structurally impossible, and a hierarchical fork-based Random Number Generator (RNG) guarantees exact reproducibility from a single integer seed. A reference 100-document corpus spans 50 enterprise domains, 4,374 pages, 16,589 QA pairs, and 21,402 assertions at zero annotation cost.

Neural circuits are infamously more robust than their deep network counterparts. Biology often explains this through degeneracy, where structurally distinct configurations support the same function. Deep learning studies a parallel phenomenon through mode connectivity, which links same-objective networks by low-loss curves, sheets, or volumes. Yet real biological systems undergo continual structural change, a more flexible form of degeneracy in which specialized circuits gradually adapt toward new competencies through localized changes while preserving existing function. The analogous question in deep learning is harder than ordinary mode connectivity, asking whether many independently initialized models trained on distinct tasks can be connected by a high-dimensional region of competent multi-task solutions. We introduce Nimbus, which learns a high-dimensional Bézier manifold connecting N specialists, going beyond the one-dimensional paths and two-dimensional sheets of prior mode connectivity work. A naive parameterization would scale combinatorially with N, but Nimbus requires only a small number of learned correction tensors shared per interaction order, showing that the connecting manifold has surprisingly low complexity yet supports multi-task competencies. The resulting manifold from Nimbus forms a navigable "Marauder's Map" connecting independently trained specialists through a shared geometry of competent solutions. Across vision (CIFAR-100) and language modeling (five disjoint text domains), Nimbus produces manifolds that are both functionally and locally weight-flat. Ensembles sampled from them degrade more gracefully than baselines under input and weight perturbations, with the largest gains in negative log-likelihood and expected calibration error. Beyond its empirical behavior, Nimbus supplies a deep learning interpretation of flexible degeneracy, recasting the connectivity between specialized circuits from a low-dimensional path between distant solutions into a high-dimensional volume of nearby ones.

GNNs have achieved remarkable success in graph learning, yet their black-box nature obscures the combinatorial reasoning behind their predictions. A core challenge lies in understanding how GNNs translate topological patterns (graph concepts) into logical rules. Current works only uncover hard Boolean logical rules over graph concepts, which cannot quantify the contribution of each concept to model predictions. Moreover, they are post-hoc methods that generate explanations after training via surrogate models, and thus may deviate from the true combinatorial reasoning of GNNs. In this work, we develop the graph concept bottleneck that enforces the combinatorial reasoning of GNNs to fit soft logical rules over graph concepts, thereby quantifying the contribution of each concept. To further enhance the graph concept bottleneck, we treat graph concepts as "graph words" and graphs as "graph sentences", and leverage language models to learn context-aware graph concept embeddings. Extensive experiments on multiple datasets show that our method GCBMs achieve state-of-the-art performance in both interpretability and classification.

Reliable evaluation is essential for progress in image generation, yet widely used metrics rely on assumptions that are misaligned with the geometry of real data. We revisit two dominant families of metrics: Fréchet-based metrics such as FID, and precision/recall-style metrics based on KNN balls. We show that FID reduces complex feature distributions to a Gaussian moment summary, making it blind to differences beyond mean and covariance. We further show that KNN-ball metrics construct inflated isotropic support estimates in high-dimensional feature spaces, causing binary membership queries to accept off-manifold samples and making recall unstable under standard finite-sample evaluations. We propose Chamfer distance as a simple alternative: instead of constructing inflated support estimates, it directly measures cross-set nearest-neighbor distances between real and generated samples. The resulting one-sided distances provide interpretable signals for fidelity and coverage, while reducing sensitivity to the neighborhood hyperparameter and remaining computationally efficient.

We introduce The FACTS Leaderboard, an online leaderboard suite and associated set of benchmarks that comprehensively evaluates the ability of language models to generate factually accurate text across diverse scenarios. The suite provides a holistic measure of factuality by aggregating the performance of models on four distinct sub-leaderboards: (1) FACTS Multimodal, which measures the factuality of responses to image-based questions; (2) FACTS Parametric, which assesses models’ world knowledge by answering closed-book factoid questions from internal parameters; (3) FACTS Search, which evaluates factuality in information-seeking scenarios, where the model must use a search API; and (4) FACTS Grounding (v2), which evaluates whether long-form responses are grounded in provided documents, featuring significantly improved judge models. Each sub-leaderboard employs automated judge models to score model responses, and the final suite score is an average of the four components, designed to provide a robust and balanced assessment of a model’s overall factuality. The FACTS Leaderboard Suite will be actively maintained, containing both public and private splits to allow for external participation while guarding its integrity. It can be found at https://www.kaggle.com/benchmarks/google/facts.

Self-attention without positional information is permutation equivariant, therefore positional encodings are required whenever a transformer must distinguish input order. However, injecting absolute position can also weaken length generalization. We formalize this tension for additive positional encodings by introducing positional spread, the total variance of the positional embedding matrix. For additive PE in softmax attention with mean pooling, we prove an upper bound demonstrating that the equivariance breaking component of risk is controlled by $\|\widetilde E\|_F^2$, along with a matching lower bound for self-separable additive PEs. We finally prove a spread-expressivity tradeoff: any additive PE that distinguishes absolute positions with minimum separation $\delta$ must have spread $\Omega(\delta^2 n)$. The theory predicts two regimes: when the target is near permutation invariant, lower spread improves length extrapolation; when the task requires absolute position, additional spread reduction destroys expressivity. We test this prediction using controlled synthetic tasks, spread-regularized additive PE, and long context retrieval and reasoning tasks. Across low positional sensitivity tasks, spread strongly predicts OOD accuracy; on absolute position tasks, performance demonstrates the predicted non-monotone tradeoff. For score-based methods such as RoPE and ALiBi, we include a displacement based comparison rather than matching bounds, clarifying both the reach and limits of the spread theory.

Memory-augmented reasoning agents typically store a single realized reasoning trajectory as the unit of experience. This abstraction is brittle under stochastic decoding: the same task can induce multiple plausible reasoning paths with different decompositions, assumptions, and failure modes. We introduce Branch-Structured Distributional Memory (BranchDIME), a framework for constructing memory from branch-structured trajectory sets rather than isolated traces. BranchDIME samples short reasoning prefixes, clusters them into prefix-induced branches, expands representative prefixes into full trajectories, and distills the resulting set into reusable memories containing positive strategies, anti-patterns, and contrastive rules. Across GPQA-DIAMOND and MATH500, BranchDIME improves over single-trajectory memory and alternative trajectory-selection baselines at comparable inference latency, yielding relative gains of 8.7% and 2.2% over single-trajectory memory, respectively. Coverage analysis shows that BranchDIME spans all discovered branches, compared with 20.0% coverage for consensus selection and 64.9% for random selection, while also achieving the highest downstream accuracy. BranchDIME further transfers across tasks and models, improving over single-trajectory memory by up to 15.4% on AIME25 and OlymMATH. These results support a shift from trajectory as experience to trajectory distribution as experience: memory should capture the local reasoning landscape induced by stochastic generation, not merely one sampled path.

Large-scale wildlife monitoring in forested environments using drones remains challenging due to occlusion, limited visibility, and scarce annotated data. We present a comprehensive airborne wildlife dataset comprising 386 paired RGB and thermal aerial video sequences recorded across diverse temperate forests and forest-adjacent habitats in Austria. Each frame is geo-referenced with precise global coordinates (longitude, latitude and altitude), enabling learning and evaluation in both image space and geographic space. The dataset includes 10 animal species captured under varying seasonal, environmental, and illumination conditions, with annotations supporting tasks such as object detection, multi-object tracking and fine-grained classification. The multispectral data and spatial metadata enable research on world-coordinate trajectory analysis, spatial population modeling, geo-aware perception, and advanced methods such as Airborne Light Field Sampling. Thermal subsets of this dataset have been used to develop and validate methods for wildlife monitoring, while the RGB data has mainly stayed untouched. To address this, we present a cross-modal pipeline to transfer thermal annotations to RGB frames next to the dataset. With this dataset we aim to provide a blueprint to promote research on multimodal, geo-referenced perception in ecology.

Layer-wise visual-text similarity in Multimodal Large Language Models (MLLMs) is widely interpreted as evidence that the language model progressively integrates visual content into a shared representation space. This reading rests on the assumption that scalar alignment scores reflect content-level cross-modal interaction. To test this assumption, we apply controlled interventions to the visual stream. Across 13 MLLMs from five families spanning 0.5B to 72B parameters, replacing projector-output visual tokens with Gaussian noise sharply reduces task accuracy, yet four standard scalar measures (CKA, SVCCA, MIR, and the leading principal-angle cosine) fail to consistently separate the corrupted stream from the original. We call this failure the \emph{alignment illusion} and trace it to the shared language-model pathway: anisotropic MLP down-projections pull visual and text tokens toward common output directions, producing \emph{weight-induced alignment}. Because this component is essentially one-dimensional, we introduce the principal-angle gap (PA gap), defined as the difference between the top two principal-angle cosines, which separates weight-induced similarity from multi-directional visual structure. Under graded visual corruption, the PA gap tracks task accuracy more consistently than the scalar scores we consider; under a structured but irrelevant image, it further exposes regimes in which internal geometry and task accuracy come apart. Internal visual-text alignment in MLLMs is therefore best read as a geometric diagnostic of the visual stream inside the language model rather than a direct proxy for content-level cross-modal interaction, and is most informative when calibrated by controlled task evidence.

Visual adversarial examples are a well-known vulnerability of deep learning (DL) systems. The emergence of vision-language models (VLMs) further expands the attack surface through multimodal interactions. Despite extensive research on adversarial defenses over the past decade, existing works on VLM robustness often overlook adaptive white-box attacks, where the adversary has full access to both the model and the defense mechanism. We show that recent defenses can be effectively bypassed under such adaptive settings and propose \method, an adaptive test-time detection method that leverages gradient information induced by a self-targeted attack maximizing the likelihood of the VLM’s generated output. Our approach consistently outperforms existing defenses across multiple victim VLMs, attack formulations, and benchmark datasets. Our code is open-source and available online.

Existing image editing methods can be generally categorized into textual instruction-based and visual prompt-based ones. Textual instructions are semantically expressive, but are limited by the coarse granularity of spatial control of the editing results. In contrast, visual prompts such as drag and point can provide precise spatial guidance, but are limited by the inherent ambiguity in semantic intent. To unify the strength of textual and visual prompts, we present Text-Vision Co-Instructed Image Editing, which jointly models textual instructions as semantic intent and sparse visual instructions as spatial guidance, aiming to achieve precise and intent-faithful image manipulation. To this end, we first construct a textual-visual instruction paired dataset with more than 23K samples derived from dynamic videos, enabling aligned supervision for cross-modal instruction. We then propose TV-Edit, a Textual-Visual instruction unified Editing framework to contextualize drag or point-based visual instructions with image-text semantics and lift them into semantic-aware control representations for pretrained editing backbones. By integrating semantic intent and spatial constraints, TV-Edit leads to more precise spatial control, less instruction ambiguity, and stronger structural consistency than text-only or drag-based alternatives. Finally, we establish TV-Edit-Bench, a deliberately designed benchmark to evaluate semantic faithfulness, spatial alignment, and visual consistency with ground-truth references and controlled textual–visual variations for reliable assessment. Our experiments across multiple editing backbones demonstrate that TV-Edit consistently yields more precise and intent-faithful edits, significantly outperforming state-of-the-art instruction-based and drag-based baselines. Data, model, and codes will be released.

Vision--language alignment powers open-vocabulary recognition, retrieval, and LVLM grounding, yet natural captions are often underspecified, making similarity brittle and overly confident under paraphrase and omitted details. We aim to learn representations whose matching is stable across caption views and whose confidence reflects how strongly text constrains an image. We propose Text as Partial Constraint (TPC), a core--residual alignment framework that treats multi-view captions as incomplete supervision: it distills a consensus semantic core as the alignment target, learns a single-view core predictor for standard inference with one query, and explicitly discourages vision--language similarity from depending on the orthogonal ``unsaid'' residual. An uncertainty-aware contrastive objective further softens alignment when caption views disagree, reducing overconfident updates under weak language constraints. Across zero-shot recognition and adversarial robustness, TPC achieves 81.42/64.05 Top-1 clean/robust accuracy on ImageNet and 76.19/52.03 on an Avg-14 transfer suite, while improving LVLM transfer with 85.16 POPE F1 and 59.57 OKVQA accuracy under an LLaVA-1.5-7B stack. These results suggest that modeling text as a partial constraint is a practical and principled route to more reliable vision--language representations under underspecified language supervision.

Training-free diffusion priors are powerful for inverse problems, but measurement guidance during reverse sampling is local: aggressive updates can improve immediate data fit while disrupting later denoising. We introduce a tethered predictive-inertial correction rule that separates proposal generation from acceptance. At each corrected step, the denoiser prediction anchors a frozen clean-space objective combining measurement fit with a noise-level-dependent tether. Heavy-ball dynamics with diffusion-scale predictive smoothing generate clean-state candidates; a separate verifier evaluates a finite dyadic set using the original unsmoothed objective, with the denoiser prediction retained as fallback. The accepted clean state is re-noised to continue sampling. The rule applies to pixel and latent diffusion with differentiable measurement operators, requires no retraining, and adds no denoiser calls inside the correction loop. We analyze when smoothing is inactive, how it changes nonlinear or decoder-composed proposal landscapes, and how objective verification yields local nonincrease, scale robustness, and displacement control. Across natural-image and accelerated MRI benchmarks, the method achieves competitive reconstruction quality with favorable speed--quality trade--offs, often reducing runtime relative to optimization-heavy diffusion solvers. On fastMRI knee reconstruction, it attains the strongest PSNR among compared methods at both acceleration factors.

We propose a test-time defense mechanism against adversarial attacks: Unlike existing methods that rely on feature filtering or smoothing, which can lead to information loss, we propose to ``combat noise with noise'' by leveraging stochastic resonance to enhance robustness while minimizing information loss. Our approach introduces small translational perturbations to the input image, aligns the transformed feature embeddings, and aggregates them before mapping back to the original reference image. This can be expressed in a closed-form formula, which can be deployed on diverse existing network architectures without introducing additional network modules or fine-tuning for specific attack types. The resulting method is entirely training-free, architecture-agnostic, and attack-agnostic. Empirically, the method achieves state-of-the-art robustness on image classification and provides the first generic test-time defense for dense prediction tasks, including stereo matching, optical flow, and monocular depth estimation. Across adversarial attacks, it recovers up to 68.1% of the accuracy loss on image classification, 71.9% on stereo matching (MAE), 29.2% on optical flow (EPE), 68.3% on depth estimation (SqRel), and 83.8% on vision-language alignment (cosine similarity).

Pre-trained vision-language models (VLMs) enable zero-shot classification by matching images with text prompts, yet their raw confidence scores are unreliable for out-of-distribution (OOD) detection. We identify a confidence bottleneck: zero-shot VLMs produce posteriors whose maximum confidence yields much weaker ID/OOD separation than supervised classifiers trained on the target label space. To theoretically explain this gap, we relate VLM confidence to a well-separated supervised reference through a Markov-kernel transformation and analyze single-threshold detection using the Youden/KS index. Our analysis shows why raw VLM confidence is unlikely to recover the clean separation achieved by supervised classifiers. Motivated by the observation that class-wise posterior distributions contain useful structure beyond maximum confidence, we propose Self-Reinforced Optimal Transport (SROT), a label-free test-time framework that reshapes VLM posteriors with optimal transport and adaptively trains a lightweight OOD detector from rectified pseudo-labels. The detector further feeds open-set evidence back into posterior alignment, forming a self-reinforcing loop between distribution-level rectification and feature-level OOD learning. Extensive experiments on standard zero-shot OOD detection benchmarks show that SROT achieves state-of-the-art performance, with ablations validating posterior alignment, adaptive detector training, and self-reinforcement.

Recent advances in generative video models have significantly improved visual realism, making detection increasingly dependent on subtle temporal inconsistencies arising from imperfect frame-to-frame coherence. However, these signals are weak and spatially localized, while most of the video content is redundant. This creates a fundamental mismatch: existing detectors rely on heavy backbones to model full videos, assuming increased capacity can capture such cues, which leads to high computational cost and limited scalability. In this work, we rethink AI-generated video detection from a representation perspective. Instead of modeling entire videos, we propose to explicitly restructure them into compact temporal slice representations that isolate informative temporal evolution. Based on this idea, we introduce a lightweight framework using frequency-aware temporal slice consistency learning, where informative spatial rows and columns are aggregated over time to form slice-domain inputs. This reparameterization suppresses redundant appearance while preserving discriminative temporal dynamics, enabling efficient and targeted detection. By constructing slices from both high- and low-frequency regions, our method exposes complementary temporal artifacts, including unstable details and abnormal motion smoothness. We instantiate this idea in a $\textit{Temporal Slice Consistency Network}$, which integrates task-driven slice localization, hierarchical positional encoding, and a lightweight Transformer to model cross-slice and cross-frequency dependencies with minimal overhead. We further introduce $\textit{AI-Artist}$, a new benchmark with artist-curated videos from recent high-fidelity generators, including Seedance and Kling. Experiments on GenVideo and AI-Artist show that our method achieves strong performance while requiring over 400× fewer FLOPs and 3.6× faster inference than the strong video-based baseline ReStraV, enabling scalable and practical deployment.

While test-time scaling has revolutionized reasoning in large language models, generative video reasoning remains bottlenecked by a single-shot paradigm. We demonstrate that searching over denoising steps cannot rescue logically flawed rollouts because spatial trajectories commit early in the diffusion process. Root-level Best-of-$N$ (BoN) sampling is similarly inefficient: reasoning errors cluster early in the temporal axis, and resampling blindly discards verified upstream progress. To unlock effective test-time scaling for video models, we introduce \textbf{Temporal Backtracking Search (TBS)}, which shifts the search space to the temporal axis. TBS transforms video generation into an iterative generate--verify--restart loop via three core mechanisms: (1) \textit{variable-$K$ conditioning} to resume generation from arbitrary clean prefixes; (2) \textit{temporal process verification} to localize failures and extract valid restart anchors; and (3) \textit{prefix-based search} to reallocate compute toward extending correct trajectories rather than root resampling. Across algorithmic, navigation, and robotics domains, TBS Pareto-dominates matched-budget BoN. In a strict out-of-distribution setting where one-shot generation collapses ($0.7\%$ for BoN), TBS achieves $22.7\%$, with every solved episode stemming from a restarted branch. Ultimately, TBS reveals that the local reasoning competence of video models far exceeds what single-shot rollouts indicate, providing a scalable test-time framework to unlock it.

Training-free conditional diffusion provides a flexible alternative to task-specific conditional model training, but existing samplers often allocate computation inefficiently: independent guided trajectories can vary widely in quality, and additional function evaluations along a single trajectory may not recover from poor early decisions. We propose **Tempered Guided Diffusion (TGD)**, an annealed sequential Monte Carlo framework for training-free conditional sampling with diffusion priors. TGD targets tempered posterior distributions over the clean signal, using noisy diffusion states only as auxiliary variables for proposing reconstructions and propagating particles. Particles are reweighted by incremental likelihood ratios, resampled, and propagated across noise levels, concentrating computation on trajectories plausible under both the prior and observation. Under idealized exact-reconstruction assumptions, full TGD yields a consistent particle approximation to the posterior as the number of particles grows. For expensive reconstruction tasks, **Accelerated TGD (A-TGD)** retains early particle exploration but prunes to a single high-likelihood trajectory partway through sampling. Experiments on a controlled two-dimensional inverse problem and image inverse problems show improved posterior approximation and favorable wall-clock speed-quality tradeoffs over independent multi-trajectory baselines.

End-to-end autonomous driving systems are known to suffer from shortcut learning and causal confusion, limiting robust generalization under distribution shift. To address this, recent Vision-Language Model (VLM) approaches introduce intermediate linguistic reasoning steps including chain-of-thought, chain-of-causation, and counterfactual (CF) reasoning, and report consistent improvements in closed-loop driving. Among these, CF reasoning has been emphasized as enabling self-reflective, high-level causal understanding, yet whether these gains reflect genuine causal reasoning or language-induced distributional bias remains unclear. In this work, we examine what CF reasoning actually induces in VLM-based autonomous driving using accident scenarios with explicit causal structure and a discrete meta-action space. We observe that CF reasoning consistently shifts the action distribution toward more conservative behaviors, accompanied by increased perplexity and policy entropy, and that this effect persists under reinforcement fine-tuning, suggesting a structural rather than transient artifact. However, visual attention analysis and causal intervention experiments show that CF reasoning does not meaningfully alter visual grounding, and VLMs respond similarly to causal and non-causal perturbations, unlike humans who selectively react to causal factors. Overall, these results suggest that CF reasoning does not induce genuine causal reasoning grounded in visual structure. Instead, it acts as a linguistic mechanism that reshapes the action distribution, teaching models what to say but not how to reason, highlighting a fundamental gap between language-based reasoning and causal understanding in VLMs.