Searcharxiv⌕ Search

arXiv subjects

Shan You

Publications and source records attributed to Shan You.

At least 19 recordsLinked to original sources
arXiv2026

StateTrace: An Object-Centric Framework for Hidden-State Spatiotemporal Reasoning in Long Videos

Existing VLMs have achieved strong performance in video understanding, yet they struggle with long-video spatiotemporal reasoning when target objects become invisible, often mistaking "invisible" for "unknown". We define this challenge as hidden-state spatiotemporal reasoning: inferring object states during prolonged invisible intervals from context interactions. To address this, we propose StateTrace, a novel object-centric framework that endows VideoLLMs with an explicit mechanism for hidden state reasoning in long videos. StateTrace builds a reusable spatiotemporal state memory that organizes object trajectories, inter-object relations, and state-transition events into a structured reasoning substrate. At inference time, it retrieves question-relevant state-evolution trajectories and converts them into compact reasoning cues, enabling the model to explicitly reason about why an object disappears, how its state evolves while invisible, and whether that state should persist at query time. We further build HSR-Bench, a diagnostic benchmark for hidden-state reasoning, containing 1,427 video-QA samples from 1,384 unique videos. Extensive experiments across multiple VideoLLMs show that StateTrace consistently improves performance on both public benchmarks and HSR-Bench (e.g., improving VideoLLaMA3 from 39.6 to 64.2 on HSR-Bench).

cs.CV↗
arXiv2026

Foresight Without Seeing: Latent Futures for World Action Models

World Action Models (WAMs) couple future visual prediction with robot action generation, enabling policies to model how the physical world evolves during interaction. Existing WAMs differ in how predictive dynamics are exposed to the action pathway. Explicit-future WAMs provide direct access to predicted scene evolution, but incur substantial inference costs from iterative video denoising. In contrast, direct-policy WAMs efficiently predict actions from the current observation but lack an explicit inference-time interface for exposing predictive dynamics to the Action DiT. To bridge this gap, we propose ForeWAM, a dynamics-conditioned direct-policy WAM that provides predictive context for action generation without decoding future videos. At its core, Future-KV performs a single Video DiT prefill over the current visual latent and stochastic future slots, and reuses the resulting layer-wise key-value states throughout action denoising. We further introduce dynamics registers supervised by a frozen latent action teacher, encouraging the implicit future states to capture interaction-induced transitions such as object motion, contact changes, and task progress. Ground-truth future observations and the teacher are used only during training; deployment requires neither and performs no future video generation. Without embodied robot data pretraining, the standard and accelerated variants of ForeWAM achieve average success rates of 96.7% and 96.9% on LIBERO, respectively. The standard variant further achieves 61.6% success on LIBERO-Plus. These results demonstrate that direct-policy WAMs can retain efficient action prediction while exposing predictive dynamics to the action pathway without explicitly generating future observations.

cs.AI↗
arXiv2026

Reward as An Agent for Embodied World Models

While RL has become a promising tool for refining world models, existing methods largely rely on conservative rollouts near the training distribution, limiting exploration, behavioral diversity, and richer dynamic discovery. In this work, we challenge this conservative paradigm. We argue that the core limitation is not exploration itself, but the lack of reliable verification strategies to support broader exploration. Without reliable verification, expanded exploration becomes highly susceptible to reward hacking, where policies exploit imperfect rewards without achieving genuine improvement. To evaluate this motivation, we instantiate our method in embodied world models, where physical plausibility, and task completion provide a rigorous testbed for scalable RL under complex dynamics. On the verification side, we introduce Reward as an Agent, an agentic reward framework that actively evaluates generated behaviors to provide robust reward signals and mitigate reward hacking under distribution shifts. On the exploration side, we introduce Dynamic-Aware Rollout Diversification through DynDiff-GRPO, which explicitly expands action-space exploration to diversify trajectories, broaden state-action coverage, and encourage richer embodied behaviors beyond conservative rollout regimes. By unifying Reward as an Agent with DynDiff-GRPO, we enable RL on a more reliable reward foundation with substantially diversified sampling, effectively mitigating reward hacking while yielding significant accuracy gains across multiple open-source world models, thereby demonstrating that broader exploration can scale successfully when grounded in robust verification.

cs.AI↗
arXiv2026

Kairos: A Regret-Aware Native World-Action Model Stack for Physical AI

We introduce \textbf{Kairos}, a regret-aware native world-action model stack for Physical AI. Kairos is motivated by the view that a physical world model should not aim to fully simulate all future pixels, but should learn and maintain the information most relevant to embodiment control: object state, spatial relations, contact conditions, task progress, action consequences, failure boundaries, and deployment uncertainty. Kairos establishes three model-side prerequisites toward this goal. First, it \textbf{learns} control-relevant information through a \textbf{Cross-Embodiment Data Curriculum}, which organizes open-world videos, human behavioral data, and robot interactions into an intervention-strength progression from passive physical observation to intentional behavior and embodied action grounding. Second, it \textbf{maintains} control-sufficient states through a unified \textbf{understanding, generation, and prediction architecture} equipped with \textbf{Hybrid Linear Temporal Attention}, where local, mid-range, and global temporal pathways support multi-timescale state maintenance under efficient inference. Third, it \textbf{deploys} these states through a \textbf{Deployment-Aware System Co-Design}, treating latency, memory footprint, and hardware compatibility as first-order constraints for future observation, action, and feedback loops. Experiments on embodied world-model benchmarks, world-action benchmarks, long-horizon generation, and inference-efficiency evaluation show that Kairos achieves superior performance while offering a favorable efficiency to capability trade-off.

cs.AI↗
arXiv2026

Seeing Realism from Simulation: Efficient Video Transfer for Vision-Language-Action Data Augmentation

Vision-language-action (VLA) models typically rely on large-scale real-world videos, whereas simulated data, despite being inexpensive and highly parallelizable to collect, often suffers from a substantial visual domain gap and limited environmental diversity, resulting in weak real-world generalization. We present an efficient video augmentation framework that converts simulated VLA videos into realistic training videos while preserving task semantics and action trajectories. Our pipeline extracts structured conditions from simulation via video semantic segmentation and video captioning, rewrites captions to diversify environments, and uses a conditional video transfer model to synthesize realistic videos. To make augmentation practical at scale, we introduce a diffusion feature-reuse mechanism that reuses video tokens across adjacent timesteps to accelerate generation, and a coreset sampling strategy that identifies a compact, non-redundant subset for augmentation under limited computation. Extensive experiments on Robotwin 2.0, LIBERO, LIBERO-Plus, and a real robotic platform demonstrate consistent improvements. For example, our method improves RDT-1B by 8% on Robotwin 2.0, and boosts $\pi_0$ by 5.1% on the more challenging LIBERO-Plus benchmark. Code is available at: https://github.com/nanfangxiansheng/Seeing-Realism-from-Simulation.

cs.CV↗
arXiv2026

Sentinel-VLA: A Metacognitive VLA Model with Active Status Monitoring for Dynamic Reasoning and Error Recovery

Vision-language-action (VLA) models have advanced the field of embodied manipulation by harnessing broad world knowledge and strong generalization. However, current VLA models still face several key challenges, including limited reasoning capability, lack of status monitoring, and difficulty in self-correction. In this paper, we introduce \textbf{Sentinel-VLA}, a metacognitive VLA model equipped with an active ``sentinel'' module to monitor real-time execution status. Only when necessary, such as during initial planning or upon detecting an error, the model triggers a dynamic reasoning or formulate error recovery solutions. This on-demand reasoning mechanism ensures robust decision-making while minimizing computational overhead. Notably, all training data (spanning 44 tasks and over 2.6 million transitions) is automatically generated and annotated through our designed pipeline. We also propose the Self-Evolving Continual Learning (SECL) algorithm, which allows Sentinel-VLA to identify its capability boundaries and automatically collect data for expansion, paired with Orthogonal Continual Adapter (OC-Adapter) to constrain parameter updates to an orthogonal space, thereby preventing catastrophic forgetting. Real-world experiments demonstrate that Sentinel-VLA boosts the task success rate by over 30\% compared to the SOTA model, PI0. We will open-source all the code, weights, and data generation pipeline.

cs.RO↗
arXiv2026

VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model

Vision-Language-Action (VLA) models have demonstrated remarkable capabilities and generalization in embodied manipulation. However, their decision-making relies on a fast, instinctive process that lacks deliberation. This strategy often leads to suboptimal or catastrophic actions when facing complex or ambiguous scenarios that require greater consideration. In this paper, we introduce \textbf{VLA-ATTC}, a framework that endows VLA models with adaptive test-time compute (TTC). VLA-ATTC employs an uncertainty-based ``cognitive clutch'' to dynamically transition from reflexive execution to a TTC deliberation phase when necessary. During TTC phase, a novel \textbf{Relative Action Critic} (RAC) model identifies the optimal action from generated candidates via pairwise comparisons. This relative mechanism replaces unstable absolute value estimation, significantly simplifying the learning objective. Furthermore, we introduce an efficient sampling strategy to amortize computational costs and an automated data pipeline that curates preference pairs without manual annotation. On the LIBERO-LONG benchmark, VLA-ATTC reduces the failure rate of the SOTA model PI0.5 by over 50\%. We will open-source all the code and weights.

cs.RO↗
arXiv2026

LEGATO: Good Identity Unlearning Is Continuous

Machine unlearning has become a crucial role in enabling generative models trained on large datasets to remove sensitive, private, or copyright-protected data. However, existing machine unlearning methods face three challenges in learning to forget identity of generative models: 1) inefficient, where identity erasure requires fine-tuning all the model's parameters; 2) limited controllability, where forgetting intensity cannot be controlled and explainability is lacking; 3) catastrophic collapse, where the model's retention capability undergoes drastic degradation as forgetting progresses. Forgetting has typically been handled through discrete and unstable updates, often requiring full-model fine-tuning and leading to catastrophic collapse. In this work, we argue that identity forgetting should be modeled as a continuous trajectory, and introduce LEGATO - Learn to ForgEt Identity in GenerAtive Models via Trajectory-consistent Neural Ordinary Differential Equations. LEGATO augments pre-trained generators with fine-tunable lightweight Neural ODE adapters, enabling smooth, controllable forgetting while keeping the original model weights frozen. This formulation allows forgetting intensity to be precisely modulated via ODE step size, offering interpretability and robustness. To further ensure stability, we introduce trajectory consistency constraints that explicitly prevent catastrophic collapse during unlearning. Extensive experiments across in-domain and out-of-domain identity unlearning benchmarks show that LEGATO achieves state-of-the-art forgetting performance, avoids catastrophic collapse and reduces fine-tuned parameters.

cs.LG↗
arXiv2025

Pistachio: Towards Synthetic, Balanced, and Long-Form Video Anomaly Benchmarks

Automatically detecting abnormal events in videos is crucial for modern autonomous systems, yet existing Video Anomaly Detection (VAD) benchmarks lack the scene diversity, balanced anomaly coverage, and temporal complexity needed to reliably assess real-world performance. Meanwhile, the community is increasingly moving toward Video Anomaly Understanding (VAU), which requires deeper semantic and causal reasoning but remains difficult to benchmark due to the heavy manual annotation effort it demands. In this paper, we introduce Pistachio, a new VAD/VAU benchmark constructed entirely through a controlled, generation-based pipeline. By leveraging recent advances in video generation models, Pistachio provides precise control over scenes, anomaly types, and temporal narratives, effectively eliminating the biases and limitations of Internet-collected datasets. Our pipeline integrates scene-conditioned anomaly assignment, multi-step storyline generation, and a temporally consistent long-form synthesis strategy that produces coherent 41-second videos with minimal human intervention. Extensive experiments demonstrate the scale, diversity, and complexity of Pistachio, revealing new challenges for existing methods and motivating future research on dynamic and multi-event anomaly understanding.

cs.CV↗
arXiv2025

Learning to Expand Images for Efficient Visual Autoregressive Modeling

Autoregressive models have recently shown great promise in visual generation by leveraging discrete token sequences akin to language modeling. However, existing approaches often suffer from inefficiency, either due to token-by-token decoding or the complexity of multi-scale representations. In this work, we introduce Expanding Autoregressive Representation (EAR), a novel generation paradigm that emulates the human visual system's center-outward perception pattern. EAR unfolds image tokens in a spiral order from the center and progressively expands outward, preserving spatial continuity and enabling efficient parallel decoding. To further enhance flexibility and speed, we propose a length-adaptive decoding strategy that dynamically adjusts the number of tokens predicted at each step. This biologically inspired design not only reduces computational cost but also improves generation quality by aligning the generation order with perceptual relevance. Extensive experiments on ImageNet demonstrate that EAR achieves state-of-the-art trade-offs between fidelity and efficiency on single-scale autoregressive models, setting a new direction for scalable and cognitively aligned autoregressive image generation.

cs.CV↗
arXiv2025

ActVAR: Activating Mixtures of Weights and Tokens for Efficient Visual Autoregressive Generation

Visual Autoregressive (VAR) models enable efficient image generation via next-scale prediction but face escalating computational costs as sequence length grows. Existing static pruning methods degrade performance by permanently removing weights or tokens, disrupting pretrained dependencies. To address this, we propose ActVAR, a dynamic activation framework that introduces dual sparsity across model weights and token sequences to enhance efficiency without sacrificing capacity. ActVAR decomposes feedforward networks (FFNs) into lightweight expert sub-networks and employs a learnable router to dynamically select token-specific expert subsets based on content. Simultaneously, a gated token selector identifies high-update-potential tokens for computation while reconstructing unselected tokens to preserve global context and sequence alignment. Training employs a two-stage knowledge distillation strategy, where the original VAR model supervises the learning of routing and gating policies to align with pretrained knowledge. Experiments on the ImageNet $256\times 256$ benchmark demonstrate that ActVAR achieves up to $21.2\%$ FLOPs reduction with minimal performance degradation.

cs.CV↗
arXiv2025

Graph Unlearning Meets Influence-aware Negative Preference Optimization

Recent advancements in graph unlearning models have enhanced model utility by preserving the node representation essentially invariant, while using gradient ascent on the forget set to achieve unlearning. However, this approach causes a drastic degradation in model utility during the unlearning process due to the rapid divergence speed of gradient ascent. In this paper, we introduce \textbf{INPO}, an \textbf{I}nfluence-aware \textbf{N}egative \textbf{P}reference \textbf{O}ptimization framework that focuses on slowing the divergence speed and improving the robustness of the model utility to the unlearning process. Specifically, we first analyze that NPO has slower divergence speed and theoretically propose that unlearning high-influence edges can reduce impact of unlearning. We design an influence-aware message function to amplify the influence of unlearned edges and mitigate the tight topological coupling between the forget set and the retain set. The influence of each edge is quickly estimated by a removal-based method. Additionally, we propose a topological entropy loss from the perspective of topology to avoid excessive information loss in the local structure during unlearning. Extensive experiments conducted on five real-world datasets demonstrate that INPO-based model achieves state-of-the-art performance on all forget quality metrics while maintaining the model's utility. Codes are available at \href{https://github.com/sh-qiangchen/INPO}{https://github.com/sh-qiangchen/INPO}.

cs.LG↗
arXiv2025

DeCoP: Enhancing Self-Supervised Time Series Representation with Dependency Controlled Pre-training

Modeling dynamic temporal dependencies is a critical challenge in time series pre-training, which evolve due to distribution shifts and multi-scale patterns. This temporal variability severely impairs the generalization of pre-trained models to downstream tasks. Existing frameworks fail to capture the complex interactions of short- and long-term dependencies, making them susceptible to spurious correlations that degrade generalization. To address these limitations, we propose DeCoP, a Dependency Controlled Pre-training framework that explicitly models dynamic, multi-scale dependencies by simulating evolving inter-patch dependencies. At the input level, DeCoP introduces Instance-wise Patch Normalization (IPN) to mitigate distributional shifts while preserving the unique characteristics of each patch, creating a robust foundation for representation learning. At the latent level, a hierarchical Dependency Controlled Learning (DCL) strategy explicitly models inter-patch dependencies across multiple temporal scales, with an Instance-level Contrastive Module (ICM) enhances global generalization by learning instance-discriminative representations from time-invariant positive pairs. DeCoP achieves state-of-the-art results on ten datasets with lower computing resources, improving MSE by 3% on ETTh1 over PatchTST using only 37% of the FLOPs.

cs.LG↗
arXiv2025

Identify, Isolate, and Purge: Mitigating Hallucinations in LVLMs via Self-Evolving Distillation

Large Vision-Language Models (LVLMs) have demonstrated remarkable advancements in numerous areas such as multimedia. However, hallucination issues significantly limit their credibility and application potential. Existing mitigation methods typically rely on external tools or the comparison of multi-round inference, which significantly increase inference time. In this paper, we propose \textbf{SE}lf-\textbf{E}volving \textbf{D}istillation (\textbf{SEED}), which identifies hallucinations within the inner knowledge of LVLMs, isolates and purges them, and then distills the purified knowledge back into the model, enabling self-evolution. Furthermore, we identified that traditional distillation methods are prone to inducing void spaces in the output space of LVLMs. To address this issue, we propose a Mode-Seeking Evolving approach, which performs distillation to capture the dominant modes of the purified knowledge distribution, thereby avoiding the chaotic results that could emerge from void spaces. Moreover, we introduce a Hallucination Elimination Adapter, which corrects the dark knowledge of the original model by learning purified knowledge. Extensive experiments on multiple benchmarks validate the superiority of our SEED, demonstrating substantial improvements in mitigating hallucinations for representative LVLM models such as LLaVA-1.5 and InternVL2. Remarkably, the F1 score of LLaVA-1.5 on the hallucination evaluation metric POPE-Random improved from 81.3 to 88.3.

cs.LG↗
arXiv2025

Learning Mask Invariant Mutual Information for Masked Image Modeling

Masked autoencoders (MAEs) represent a prominent self-supervised learning paradigm in computer vision. Despite their empirical success, the underlying mechanisms of MAEs remain insufficiently understood. Recent studies have attempted to elucidate the functioning of MAEs through contrastive learning and feature representation analysis, yet these approaches often provide only implicit insights. In this paper, we propose a new perspective for understanding MAEs by leveraging the information bottleneck principle in information theory. Our theoretical analyses reveal that optimizing the latent features to balance relevant and irrelevant information is key to improving MAE performance. Building upon our proofs, we introduce MI-MAE, a novel method that optimizes MAEs through mutual information maximization and minimization. By enhancing latent features to retain maximal relevant information between them and the output, and minimizing irrelevant information between them and the input, our approach achieves better performance. Extensive experiments on standard benchmarks show that MI-MAE significantly outperforms MAE models in tasks such as image classification, object detection, and semantic segmentation. Our findings validate the theoretical framework and highlight the practical advantages of applying the information bottleneck principle to MAEs, offering deeper insights for developing more powerful self-supervised learning models.

cs.CV↗
arXiv2024

MetaDD: Boosting Dataset Distillation with Neural Network Architecture-Invariant Generalization

Dataset distillation (DD) entails creating a refined, compact distilled dataset from a large-scale dataset to facilitate efficient training. A significant challenge in DD is the dependency between the distilled dataset and the neural network (NN) architecture used. Training a different NN architecture with a distilled dataset distilled using a specific architecture often results in diminished trainning performance for other architectures. This paper introduces MetaDD, designed to enhance the generalizability of DD across various NN architectures. Specifically, MetaDD partitions distilled data into meta features (i.e., the data's common characteristics that remain consistent across different NN architectures) and heterogeneous features (i.e., the data's unique feature to each NN architecture). Then, MetaDD employs an architecture-invariant loss function for multi-architecture feature alignment, which increases meta features and reduces heterogeneous features in distilled data. As a low-memory consumption component, MetaDD can be seamlessly integrated into any DD methodology. Experimental results demonstrate that MetaDD significantly improves performance across various DD methods. On the Distilled Tiny-Imagenet with Sre2L (50 IPC), MetaDD achieves cross-architecture NN accuracy of up to 30.1\%, surpassing the second-best method (GLaD) by 1.7\%.

cs.CV↗
arXiv2024

Decoupled Video Generation with Chain of Training-free Diffusion Model Experts

Video generation models hold substantial potential in areas such as filmmaking. However, current video diffusion models need high computational costs and produce suboptimal results due to extreme complexity of video generation task. In this paper, we propose \textbf{ConFiner}, an efficient video generation framework that decouples video generation into easier subtasks: structure \textbf{con}trol and spatial-temporal re\textbf{fine}ment. It can generate high-quality videos with chain of off-the-shelf diffusion model experts, each expert responsible for a decoupled subtask. During the refinement, we introduce coordinated denoising, which can merge multiple diffusion experts' capabilities into a single sampling. Furthermore, we design ConFiner-Long framework, which can generate long coherent video with three constraint strategies on ConFiner. Experimental results indicate that with only 10\% of the inference cost, our ConFiner surpasses representative models like Lavie and Modelscope across all objective and subjective metrics. And ConFiner-Long can generate high-quality and coherent videos with up to 600 frames.

cs.CV↗
arXiv2024

LocalMamba: Visual State Space Model with Windowed Selective Scan

Recent advancements in state space models, notably Mamba, have demonstrated significant progress in modeling long sequences for tasks like language understanding. Yet, their application in vision tasks has not markedly surpassed the performance of traditional Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). This paper posits that the key to enhancing Vision Mamba (ViM) lies in optimizing scan directions for sequence modeling. Traditional ViM approaches, which flatten spatial tokens, overlook the preservation of local 2D dependencies, thereby elongating the distance between adjacent tokens. We introduce a novel local scanning strategy that divides images into distinct windows, effectively capturing local dependencies while maintaining a global perspective. Additionally, acknowledging the varying preferences for scan patterns across different network layers, we propose a dynamic method to independently search for the optimal scan choices for each layer, substantially improving performance. Extensive experiments across both plain and hierarchical models underscore our approach's superiority in effectively capturing image representations. For example, our model significantly outperforms Vim-Ti by 3.1% on ImageNet with the same 1.5G FLOPs. Code is available at: https://github.com/hunto/LocalMamba.

cs.CV↗