NeurIPS 2026 论文列表

Judging image quality is not only ecologically relevant to everyday humans tasks, but also underpins many machine vision tasks such as image generation. This paper proposes a framework to understand the inherent perceptual space underlying image quality judgment in humans. We propose a multi-dimensional observer model that represents images as distributions in a latent perceptual space and that models human judgment as comparing noisy samples. Being constrained by neural representations in the primate ventral stream and fit to large-scale behavioral data, the model enables analysis of perceptual structure while matching the predictive power of existing metrics. Using this model, we find that the perceptual spaces needed to account for image quality judgment in humans are extremely low-dimensional compared to the image space even when considering its sparsity. The exact structure of the space (e.g., dimensionalities, information encoded) varies between low-level and high-level quality judgments, suggesting that, despite a shared retinal encoding in the beginning, humans selectively construct task-dependent perceptual spaces in visual decision making.

A striking case of goal misgeneralization was previously observed in OpenAI's Minecraft agent VPT: it killed villagers standing under some leaves, mistaking them for trees. Although this agent was released publicly, enabling white-box interpretability research, few open-weight model organisms of misalignment exist outside the LLM space. In this work, we release MultiSTEVE-1s -- a model zoo of 140 fine-tuned versions of VPT, over 1,000 training checkpoints, and an interpretability suite for analysing them. Specifically, we use the STEVE-1 training procedure to add instruction-following capabilities to VPT with fixed hyperparameters and controlled variations in training randomness. We demonstrate the utility of MultiSTEVE-1s by showcasing the research it enables. First, for some training runs, the only difference is a least-significant bit flip in a single initialised weight. Others differ in the full randomness for weight initialisation and data. Yet, the single bit-flip setting produces agents that act nearly as differently from each other as the full randomness ones. Second, we use our interpretability suite to show that several known VPT attention heads retain their roles after STEVE-1 fine-tuning, while attention strength to the same behaviourally meaningful frame can vary substantially across agents and checkpoints. Finally, even though the agents are similarly capable in in-distribution tasks, the out-of-distribution behaviour of villager killing can differ substantially between them -- in one setting, an agent kills villagers less than 5\% of the time, while another agent kills them nearly 50\% of the time. Our results show the value of studying multiple similarly trained agents, rather than acting like a behavioural biology lab with only one rat.

Tabular Foundation Models have recently established the state of the art in supervised tabular learning, by leveraging pretraining to learn generalizable representations of numerical and categorical structured data. However, they lack native support for unstructured modalities such as text and image, and rely on frozen, pretrained embeddings to process them. On established Multimodal Tabular Learning benchmarks, we show that tuning the embeddings to the task improves performance. Existing benchmarks, however, often focus on the mere co-occurrence of modalities; this leads to high variance across datasets and masks the benefits of task-specific tuning. To address this gap, we introduce MulTaBench, a benchmark of 40 datasets, split equally between image-tabular and text-tabular tasks. We focus on predictive tasks where the modalities provide complementary predictive signal, and where generic embeddings lose critical information, necessitating Target-Aware Representations that are aligned with the task. Our experimental results demonstrate that the gains from target-aware representation tuning generalize across both text and image modalities, several tabular learners, encoder scales, and embedding dimensions. MulTaBench constitutes the largest image-tabular benchmarking effort to date, spanning high-impact domains such as healthcare and e-commerce. It is designed to enable the research of novel architectures which incorporate joint modeling and target-aware representations, paving the way for the development of novel Multimodal Tabular Foundation Models.

Motion instruction generation in cross-video comparison aims to produce corrective feedback that describes the differences between a query and a reference motion. However, existing models often generate instructions that exhibit motion hallucinations, failing to reflect actual kinematic differences between paired videos. To systematically investigate these hallucinations, we introduce MotionHalluc, a dedicated benchmark for evaluating motion hallucinations in paired-video comparison. MotionHalluc comprises 1540 fine-grained questions over 553 video pairs, evaluating hallucinations along three core dimensions: (1) directional hallucination, (2) attributional hallucination, and (3) temporal hallucination. Extensive evaluations of state-of-the-art large multimodal models demonstrate high susceptibility to these hallucinations. Furthermore, we provide Perceive-Parse-Verify (PPV) as a training-free measurements extraction and verification baseline that converts candidate instructions into executable measurement queries and supplies kinematic measurements at inference time. Our results show that this simple measurements injection yields an average 10.6% performance gain across models, suggesting that motion reasoning with explicit quantitative measurements is a key factor in reducing hallucinations in cross-video comparison.

Despite recent advancements in Text-to-Video (T2V) synthesis, generating high-fidelity and dynamic motion remains a significant challenge. Existing methods primarily rely on Classifier-Free Guidance (CFG), but standard CFG treats all semantic dimensions uniformly and provides no mechanism to selectively resolve ambiguity along the motion axis. To address this, we propose MotionCFG, a training-free approach that introduces selective guidance along the motion subspace of the condition embedding. Specifically, we construct a motion-perturbed negative condition by injecting Gaussian noise into motion-related embeddings, and steer generation away from it to selectively amplify motion intent. We show theoretically that this procedure implicitly approximates *Semantic Laplacian Sharpening*, a second-order correction that suppresses motion-ambiguous regions of the score landscape and amplifies well-resolved dynamic peaks. Combined with a piecewise guidance schedule that confines intervention to the early denoising steps, MotionCFG consistently improves motion dynamics across state-of-the-art T2V frameworks with negligible overhead. We further demonstrate that this Laplacian sharpening principle generalizes beyond motion, effectively steering complex, non-linear concepts such as precise object numerosity that are typically difficult to modulate via standard text-based guidance.

LiDAR scene flow estimation is essential for autonomous driving, as it provides 3D motion for each point. Self-supervised approaches use static-dynamic classification to mitigate the imbalance between static and dynamic points, deriving targeted supervision. However, existing methods rely on sparse geometric observations for this classification, making them vulnerable to data sparsity and occlusions. The resulting noisy labels provide incorrect motion guidance and degrade scene flow learning. To address this, we introduce TrackCue, a tracking-guided framework for improving dynamic object representation in LiDAR scene flow estimation. In particular, TrackCue repurposes point tracking to obtain dense image-space trajectories anchored to LiDAR points, providing motion cues beyond sparse geometric observations. Furthermore, we present a visually consistent motion compensation strategy that compares the tracked trajectories with ego-induced rigid trajectories in the image plane, effectively isolating true object motion from ego-induced apparent motion. To transfer these isolated motion cues back to the LiDAR domain, we perform visual motion cue lifting, which associates ego-compensated image trajectories with LiDAR points for static-dynamic label refinement. As a result, TrackCue produces more accurate static-dynamic classification and provides more reliable supervision for scene flow learning. Experimental results show that TrackCue significantly improves the precision and F1 score of dynamic labels, leading to performance gains in self-supervised scene flow estimation.

Microscopy images contain rich information about how cells respond to perturbations, making them essential to applications like drug screening. To quantify images, researchers often use representation extraction methods, and recent years have seen a proliferation of deep learning methods. While measuring the quality of these representations is essential, evaluation remains fragmented, with each proposed model evaluated on different tasks and datasets, using custom pipelines and metrics, making it difficult to fairly compare models. Here, we introduce MorphoHELM, a comprehensive open benchmark for evaluating feature extraction methods for cell painting, the most widely-used morphological profiling assay. MorphoHELM consolidates evaluation standards in the field, extends and corrects them to be more robust, and evaluates on the widest range of methods to date. A defining feature of the benchmark is that each task is evaluated at different degrees of batch effects (or technical noise), directly quantifying how the ability of methods to detect biological signal degrades as noise increases. Together, these properties enable MorphoHELM to detect trade-offs between methods, and we demonstrate that models that excel at certain kinds of biological signal are weaker at others. We show that no existing model outperforms classic computer vision analytic strategies across all settings, which remain the strongest general use-case representations. All datasets, code, and evaluation tools are publicly available.

Vision-Language Model (VLM) driving agents promise explainable end-to-end autonomy by first producing natural-language reasoning and then predicting trajectory planning. However, whether planning is **causally** driven by this reasoning remains a critical but unverified assumption. To investigate this, we build DriveMind, a large-scale driving Visual Question Answering corpus with plan-aligned Chain-of-Thought (CoT), automatically generated from nuPlan. Our data generation process converts sensors and annotations into structured inputs and, crucially, separates priors from to-be-reasoned signals, enabling clean information ablations. Using DriveMind, we train representative VLM agents with Supervised Fine-Tuning and Group Relative Policy Optimization and evaluate them with nuPlan’s metrics. Our results, unfortunately, indicate a consistent **causal disconnect** in reasoning-planning: removing ego/navigation priors causes large drops in planning scores, whereas removing CoT produces only minor changes. Attention analysis further shows that planning primarily focuses on priors rather than the CoT. Based on this evidence, we propose the Reasoning-Planning Decoupling Hypothesis, positing that the training-yielded reasoning is an ancillary byproduct rather than a causal mediator. To enable efficient diagnosis, we introduce a novel, training-free probe that measures an agent's reliance on priors by evaluating its planning robustness against minor input perturbations. In summary, we provide the community with a new dataset and a diagnostic tool to evaluate the **causal fidelity** of future models.

The open-source model ecosystem now contains hundreds of thousands of pretrained models, yet picking the best model for a new dataset is increasingly infeasible: new models and unbenchmarked datasets emerge continuously, leaving practitioners with no prior records on either side. Existing approaches handle only fragments of this in-the-wild setting: AutoML and transferability estimation select models from small predefined pools or require expensive per-model forward passes on the target dataset, while model routing presupposes a given candidate pool. We introduce ModelLens, a unified framework for model recommendation in the wild. Our key insight is that public leaderboard interactions, though scattered and noisy, collectively trace out an implicit atlas of model capabilities across heterogeneous evaluation settings, a signal rich enough to learn from directly. By learning a performance-aware latent space over model--dataset--metric tuples, ModelLens ranks unseen models on unseen datasets without running candidates on the target dataset. On a new benchmark of 1.62M evaluation records spanning 47K models and 9.6K datasets, ModelLens surpasses baselines that either rely on metadata alone or require running each candidate on the target dataset. Its recommended Top-K pools further improve multiple representative routing methods by up to 81\% across diverse QA benchmarks. Case studies on recently released benchmarks confirming generalization to both text and vision-language tasks.

Multimodal dataset distillation (MDD) seeks to synthesize compact surrogates from large-scale image-text corpora, reducing the high cost of pre-training. Although recent feature-matching methods have improved the efficiency of MDD, they often suffer from initialization overfitting, which distorts the cross-modal structure and severely degrades retrieval performance under unseen initializations. To mitigate this issue, we propose $\textbf{MDA-DD}$, which enhances generalization via a dynamic staggered model queue. We further identify two architectural bottlenecks in prior work: structurally redundant projections and inefficient covariance computation. For the former, we design an asymmetric dual projection to reduce redundancy while preserving balanced bidirectional cross-modal alignment. For the latter, we introduce a Second-Order Cross-Modal Moment (SOCM) objective, which efficiently captures both mean and correlation statistics across modalities. Experiments on Flickr30K and MS-COCO show that $\textbf{MDA-DD}$ consistently surpasses existing methods, yielding notable retrieval improvements (e.g., +12.9 IR@1 and +18.2 TR@1 in the 100-pair setting) and matching or exceeding baselines trained on 500 pairs using only 100 pairs.

Mixture-of-Experts (MoE) enables multimodal large language models (MLLMs) to scale capacity efficiently, but expert parameters dominate memory—accounting for over 90% of model size in modern MoE-MLLMs (e.g., 54 GB out of ∼58 GB in BF16 for InternVL3.5-30B-A3B and Qwen3-VL-30B-A3B)—even though only a few experts are activated per token. Expert pruning has emerged as an effective compression strategy, but existing methods aggregate importance uniformly across tokens, ignoring that the same expert may serve different roles for vision versus language processing. To address this, we propose MAEP (Modality-Aware Expert Pruning), a training-free framework that decomposes expert importance by modality in a single forward pass. MAEP introduces (1) Delta-H Output (∆H), a redistribution-aware metric measuring the MoE output change when an expert is removed, and (2) low-attention vision tokens as a vision-side pruning signal that is dataset-stable: its expert-importance ranking remains stable across different datasets. Text-side importance serves as a preservation signal while vision-side importance serves as a pruning signal, and experts are ranked globally across layers. Experiments on three MoE-based MLLMs across multimodal and text benchmarks show that MAEP outperforms modality-agnostic baselines on the majority of multimodal benchmarks while remaining competitive on text-only tasks, reducing the BF16 expert weight footprint by ∼27 GB (∼47% of total memory) so that the expert-weight footprint fits within a single 40 GB GPU.

Despite significant advances in large vision-language models (Video-LLMs) for general video understanding, accurately narrating fine-grained, highly dynamic human activities remains a formidable challenge. Existing approaches typically rely on global sequence embeddings or isolated part-level retrieval, lacking the precise relational kinematic grounding required to prevent physical hallucinations. To address this, we introduce Motion Retrieval-Augmented Generation for Detailed Video Captioning (MoRe-DVC). Our algorithmic novelty centers on two core contributions: an event-boundary-aware graph-conditioned retrieval mechanism that captures hierarchical and chronological motion dependencies, and a symbolic physical faithfulness reward that explicitly constrains generation to grounded physical reality. Specifically, MoRe-DVC integrates a Pose-Orientation Encoding Module (POEM) to extract symbolic descriptors, an Event Calibration and Refinement Module (ECRM) to dynamically adjust coarse action boundaries into an event graph, and a Graph-Conditioned Hybrid Retriever (GCHR) to select relationally relevant evidence. By coupling graph-level retrieval with a strict symbolic reward, our framework reduces motion-language inconsistency. Extensive experiments demonstrate that MoRe-DVC achieves robust outperformance across three motion-centric datasets (BoFiT, HumanML3D, and FineMotion), establishing a new standard for physically faithful action narration.

Mixture-of-Experts (MoE) language models scale model capacity efficiently by activating a small subset of parameters per token. However, in long-context inference, memory remains a major bottleneck because the KV cache grows linearly with sequence length. Existing KV cache compression methods reduce this cost for dense transformers, but overlook a problem specific to MoE: KV eviction can change which experts are selected during answer generation, degrading the generation quality. We propose MoEZip, a routing-aware KV cache compression for sparse MoE language models. To estimate the sensitivity of expert routing to perturbation, we propose a novel metric based on the Fisher information matrix of routing distributions. MoEZip combines this metric with attention weights for routing-aware scoring of contexts. Finally, we compose a complementary scoring system that combines eviction methods for dense and sparse-MoE architectures. Experiments show that MoEZip achieves robust performance under aggressive cache compression on various benchmarks. At a KV retention ratio of 10\%, MoEZip improves over the SoTA baseline by +24 and +20 points on average on LongBench and RULER, corresponding to 2.3$\times$ and 4.9$\times$ higher scores, respectively.

Existing Mamba-based diffusion backbones struggle to achieve competitive performance in vision tasks, as purely sequential propagation lacks an explicit spatial inductive bias and relies on a fixed computation pattern across timesteps. We address this limitation by proposing \textbf{Mixture-of-Hierarchical Experts (MoH)}, a Mamba-based diffusion backbone that performs timestep-conditioned routing over architectural components. MoH dynamically selects representation levels by its hierarchical expert space, adapts feature transformations, and adjusts Mamba scan depth according to the denoising state, enabling flexible modeling of spatial structure and multi-scale dependencies. This design leads to substantial performance gains. On unconditional CelebA-HQ $256 \times 256$, MoH reduces FID from 14.27 to 7.99, and on MS-COCO, it improves FID from 41.80 to 20.00, significantly outperforming prior Mamba-based models while improving training efficiency. These results suggest that routing over architectural components can provide an effective mechanism for improving Mamba-based diffusion backbones.

Multimodal Large Language Models (MLLMs) are typically designed under the assumption that all modalities available during training will also be accessible at inference. However, many real-world settings violate this assumption, requiring models to operate under a privileged modality setting, where auxiliary modalities are available only during training. While these modalities contain valuable information, existing MLLMs largely fail to leverage them effectively, as they treat modalities as interchangeable inputs rather than sources of complementary supervision. We propose Mixture of Probes (MoP), a novel framework that disentangles modality-specific and modality-general signals within the MLLM, allowing the model to preserve modality-dependent structure while learning transferable representations across modalities. At its core, MoP achieves this through a structured probing mechanism that extracts and organizes information from intermediate representations of a shared modality encoder, rather than relying only on final-layer alignment as done in existing MLLMs. To support this disentanglement, we further introduce MoP Cross-modal Probe Training (MoP-X), a training strategy for MoP centered around a probe disentanglement loss that prevents probe collapse and encourages cross-modal learning. We evaluate MoP across two domains spanning eight tasks and four modalities under a comprehensive evaluation protocol tailored to the privileged modality setting, where each modality is independently treated as the sole input at inference time. MoP consistently outperforms strong MLLM baselines, achieving up to 65% relative improvement, demonstrating that auxiliary modalities, even when unavailable at inference, can provide substantial gains when effectively leveraged during training.

Pre-trained vision encoders contain layer-wise visual representations that differ in spatial granularity, semantic abstraction, and sensitivity to local details. However, most Multimodal Large Language Models (MLLMs) rely only on the final or penultimate vision encoder representations or fixed aggregation rules, making visual abstraction largely query-agnostic and limiting access to fine-grained cues such as small objects, spatial details, text, and subtle visual attributes. In this work, we propose **Mixture of Layers (MoL)**, an instruction-conditioned layer routing approach that dynamically aggregates query-relevant latent representations from intermediate vision encoder layers. Given a text query, **MoL** predicts routing probabilities over vision encoder layers and aggregates selected hidden states at either the image level, patch level, or through a hybrid routing mechanism. Our proposed **MoL** is a vision encoder-agnostic framework for layer-wise routing that selectively samples visual features useful for fine-grained visual reasoning tasks. Experiments across seven fine-grained visual reasoning benchmarks demonstrate substantial performance improvements, including +18.9% on V* overall accuracy, +4.5% on HRBench4K, and +16.3% on CharXiv compared to baseline MLLMs, *without* requiring multi-resolution inputs, simple interleaving of multiple vision encoders, or additional patch tokens. We further analyze receptive field scales and routing behaviors across vision encoder layers to explain why adaptive layer selection improves perception and reasoning. Our results suggest that conditional intermediate-layer representations are a key step toward stronger visual perception and reasoning in MLLMs. Code will be open-sourced.

Neural Combinatorial Optimization (NCO) is a promising paradigm for solving routing problems, yet its generalization across diverse data distributions remains a critical bottleneck. Existing methods typically tackle cross-distribution generalization through complex multi-distribution training or intricate meta-learning schemes, implicitly presuming that single-distribution training yields a weak zero-shot baseline. In this paper, we demonstrate that the potential of single-distribution training has been substantially underestimated and can be unlocked by a simple architectural inductive bias: Mix Normalization. To approach this, we first construct a comprehensive benchmark spanning three representative routing problems and 179 fine-grained datasets, capturing diverse shifts in node coordinates, customer demands, and time windows. Building on this benchmark, we propose MixRoute, a neural framework that adaptively combines normalization statistics at different granularities to mitigate varying distribution shifts. Trained exclusively on uniform distributions, MixRoute generalizes effectively to a wide range of unseen distributions without any distribution-specific adaptation. Extensive experiments demonstrate that MixRoute consistently achieves state-of-the-art zero-shot generalization. Our work revisits normalization as a parsimonious alternative for zero-shot generalization against complex training-adaptation schemes.

Detecting AI-generated videos (AIGV) in a generator-agnostic manner is increasingly important as generative models close the gap with real footage, and a video, with many frames, intuitively offers richer forensic evidence than a single image. Yet existing detectors fall short of this expectation: under the encode-then-fuse paradigm, increasing the frame budget from 1 to 8 improves a voting-based image detector by only 5.28%, and even temporal-modeling backbones such as TimeSformer barely improve on this. Per-frame predictions are also noisy: per-frame logits fluctuate substantially within a video and correlate weakly with any frame-level proxy. We trace the cause not to the aggregation step but to per-frame encoding severing multi-frame cues before they can interact, leaving any post-hoc selection, weighting, or aggregation strategy without a reliable basis. We therefore propose **MixForensics**, a blend-then-encode framework that combines multiple frames in the pixel domain into a single composite image (Stochastic Frame Blending, SFB) and regularizes the encoder for consistency across blended views (Blend Invariance Regularization, BIR). To stress-test this approach, we further introduce **MixForensics-Bench**, a test-only benchmark covering ten recent generators (including five closed-source commercial platforms) and two partial-forgery scenarios (temporal splicing and conditional continuation), with matched controls that isolate generative artifacts from editing artifacts. On AIGVDBench and MixForensics-Bench, MixForensics outperforms the strongest baseline by 1.27% and 8.11% on the fully-fake splits, leads on both partial-forgery scenarios by 11.49% on temporal splicing and 9.15% on conditional continuation, and uses fewer encoder forward passes than per-frame methods.

When learning new concepts sequentially, Vision-Language Models (VLMs) face forgetting that degrades both previously acquired tasks and inherent zero-shot capabilities. This challenge is particularly pronounced in the Cross-domain Task-Agnostic Incremental Learning (X-TAIL) setting, where the absence of explicit task identifiers forces diverse concepts to coexist within a unified and crowded multi-modal space. In this context, we identify an underexplored vulnerability termed Asymmetric Boundary Encroachment (ABE). Even when historical feature spaces remain stable, unconstrained new text features actively encroach upon these established sub-spaces during training. To systematically counteract ABE, we propose a unified framework. First, we introduce Analytic Decision Boundary (ADB), which constructs a geometric defense by enforcing an analytic margin derived from cumulative statistics to secure historical multi-modal boundaries. Furthermore, we integrate Orthogonal Visual Adaptation (OVA) and Analytic Ridge Classifier (ARC) to safely evolve the visual backbone and enhance joint inference. Experiments under X-TAIL setting demonstrate that our framework effectively mitigates ABE and consistently achieves new state-of-the-art performance. Our code is available at https://anonymous.4open.science/r/abe-cl-vlm.