We introduce 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 learns control-relevant information through a 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 maintains control-sufficient states through a unified understanding, generation, and prediction architecture equipped with Hybrid Linear Temporal Attention, where local, mid-range, and global temporal pathways support multi-timescale state maintenance under efficient inference. Third, it deploys these states through a 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.

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.

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).

Marine heatwaves (MHWs) are prolonged periods of anomalously warm ocean temperatures that threaten marine ecosystems and regional economies. Reliable and efficient forecasting with local details across multiple temporal scales is crucial to mitigate their negative impacts. However, current dynamical methods are computationally intensive and struggle to capture local stochastic climate anomalies, while statistical methods fail to explicitly model atmospheric interactions governing MHW evolution. We present MARINA, a multi-temporal resolution forecasting model that integrates physical insights into a statistical framework, enabling skillful and efficient MHW forecasting. To effectively train MARINA, we built MT-MHW, a multi-temporal resolution MHW dataset comprising 3.22 million data points and incorporating key meteorological variables from multiple weather stations. Together, MT-MHW enables MARINA to model the complex interactions among meteorological factors that govern MHW evolution across multiple timescales in regional areas. MARINA facilitates the low-cost integration of dynamical knowledge into statistical models, enabling detailed MHW forecasting across previously unavailable temporal scales and enhancing region-specific disaster prevention.

In recent years, text-to-image generation has witnessed rapid advancements, among which personalized facial generation has garnered significant attention. This technology has been widely applied in areas such as virtual avatar construction and creative content generation. However, existing methods exhibit poor performance when generating face images for minority or underrepresented race-gender groups. They often result in feature entanglement, where non-target race-gender attributes are fused into the output, leading to distortion, identity loss, and fairness issues. To address this challenge, we propose FairMoE, a Mixture of Experts (MoE) framework assigning decoupled experts to each race-gender category. FairMoE stores knowledge for various groups in different experts and dynamically routes to the appropriate expert according to the generation goal. This approach, by decoupling knowledge, effectively mitigates racial feature entanglement. Additionally, we develop AttrCtrl, a structured attribute controller optimized through cyclic feedback, which can provide fine-grained semantic control. Extensive experiments demonstrate that our method achieves superior attribute diversity, semantic alignment, and fairness, outperforming existing SOTA methods across multiple benchmarks. Specifically, it achieves a CLIP score of 0.2602, which is a 34.2

Federated Learning (FL) faces a fundamental dilemma: existing defenses against gradient leakage attacks (GLAs) invariably sacrifice model performance for privacy protection through noise injection or gradient clip. We introduce Federated Learning with Momentum-Based Orthogonal Projection (FedMOP), a method that simultaneously achieves strong privacy guarantees and superior model performance. The key insight is to leverage initialization-based offset mechanisms that operate on orthogonal dimensions. For performance enhancement, FedMOP employs gradient orthogonal projection to counteract local drift, effectively offsetting each client's round-training initial model using global statistical context. For privacy protection, it introduces momentum-based trajectory offset hiding, which makes the offset vector inherently unrecoverable by constructing information barriers through private initialization and randomized evolution. These two mechanisms are synergistic rather than antagonistic. Theoretically, we prove convergence preservation and characterize the computationally infeasible inverse problem faced by attackers. Extensive experiments on CIFAR-10/100 and Tiny-ImageNet demonstrate that FedMOP not only defends effectively against state-of-the-art GLAs but also surpasses existing FL methods in both accuracy and convergence speed, validating its ability to jointly enhance privacy and performance in FL.

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

Vision-Language Models (VLMs) have demonstrated powerful generalization in open-vocabulary object detection. However, existing VLMs struggle with robustly recognizing concepts at varying granularities (e.g., "animal" vs. "dog"), stemming from their flat, non-hierarchical pre-training objectives. A prevailing paradigm attempts to mitigate this by injecting hierarchical information to expand cross-granularity semantic matching. While effective, our work reveals that this approach often disrupts generalization on coarse-grained concepts and fails to correct the models’ inherent pre-training biases. To address this, we introduce the Intrinsic Semantic Consistency Enhancement (InCoe) framework. By distilling universal core semantics to mitigate granularity conflicts and performing cross-modal debiasing with pre-computed priors, InCoe achieves robust and consistent representation alignment across all granularity levels. Extensive experiments on diverse benchmarks show InCoe not only resolves the performance degradation on general concepts but also consistently outperforms state-of-the-art methods, achieving up to 7.1% relative improvements in mAP50 over SHiNe.

Vision-Language Models (VLMs) are widely used in tasks like Open-Vocabulary Object Detection and zero-shot Classification, owing to their powerful generalization. However, recent research reveals that VLMs exhibit significant performance instability when tasked with recognizing concepts at varying granularities (e.g., ``animal'' vs. ``dog''). Prevailing methods inject external knowledge from Large Language Models, but this unconstrained approach distorts the VLM's inherent hierarchical orthogonal geometry, leading to performance collapse on general concepts. To address this, we introduce GeCoin, an innovative Geometrically Constrained framework that safely enhances existing VLMs with external knowledge for improved hierarchical understanding, without additional training. By projecting knowledge into the null-space of a query concept's feature space, GeCoin mathematically guarantees the preservation of general knowledge while integrating specialized information. Extensive experiments across large-scale benchmarks, diverse VLMs, and knowledge from various LLMs (e.g., GPT-3.5, Claude-3, Gemini-Pro) show that GeCoin boosts performance by an average of 3.9% over the strongest baseline—crucially eradicating performance collapse on general concepts.

Partial domain adaptation remains challenging when label spaces are asymmetric: private source classes dominate alignment, while adaptation signals are miscalibrated across different model views. We argue two under-explored factors jointly drive failure: (i) insufficient controllability over transfer selection under label asymmetry, and (ii) miscalibration between geometric similarity and classifier logits, yielding decision conflicts and noisy-target amplification. In this paper, we present P2CL, a Prototype-constrained consistent Learning framework with Controllable Prototype-gated Transfer to decide what and how to align, prioritizing highly transferable samples and suppressing private-class interference. To reduce decision conflicts, we introduce Discrepancy-gated Consistency Calibration, a lightweight mechanism aligning geometric cues with classifier logits via discrepancy-aware consistency constraints, suppressing noise amplification. Extensive experiments across three benchmarks show consistent gains over recent strong approaches and clearer shared-class separation in adaptation.

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\%.

Robotic manipulation in complex 3D environments requires unifying spatial reasoning with intuitive visual perception, which is a capability that current Vision-Language-Action paradigms address separately. While 3D VLAs excel in geometric and physical reasoning, they lack intuitive, image-level understanding and dense visual semantics; conversely, 2D VLAs (even with depth image) provide rich visual intuition and semantic continuity but miss explicit spatial global grounding. We introduce DiffRender-VLA, a differentiable rendering–based framework that bridges 3D and 2D Vision-Language-Action models through gradient-consistent visual mediation. It generates differentiable images by localizing the next end-effector target with a world-aligned cube marker, differentiably structuring surrounding geometry whose color encodes spatial relations to the marker, and rendering adaptive viewpoints optimized to reveal the target–environment spatial relationships. These differentiable images serve as visual bridges, embedding spatial semantics while allowing gradients from 2D VLAs to backpropagate into 3D representations, thereby coupling geometric reasoning with visual perception. This closed differentiable loop unifies reasoning and perception, substantially improving performance under occlusion, clutter, and complex spatial manipulation tasks, achieving average improvements of +12.1% over state-of-the-art methods.

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.

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.

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

Dynamic novel view synthesis (NVS) aims to render time-varying scenes from arbitrary viewpoints, balancing rendering quality and computational efficiency. While recent 4D Gaussian Splatting approaches offer promising real-time performance, they fundamentally overlook critical interdependence between Gaussians by modeling deformations independently. Our information-theoretic analysis reveals substantial mutual information across the Gaussian field, manifesting as appearance-preserving radiance coherence and motion-consistent deformation propagation. This finding establishes that rendering quality emerges from coordinated transformation rather than independent processing. We propose Correlationaware Dynamic Gaussian Splatting (CaDGS) with our novel Gaussian Correlation Tensor Projection (GCTP) method, which efficiently transforms the complex O(n(3)) mutual information tensor into a dual-channel O(n(2)) spatial matrix, preserving the critical topological structure of Gaussian interactions. Combined with our SpatioTemporal Deformation Consistency (STDC) learning, which enforces volumetric coherence through tensor-guided regularization across multiple scales, CaDGS prevents geometric distortions and texture inconsistencies common in previous approaches. Experimental results demonstrate state-of-the-art performance, achieving 32.4 PSNR on the Neu3D dataset with fewer Gaussians while maintaining rendering speeds of 323 FPS at 1353 x 1014 resolution.

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

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.

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 SElf-Evolving Distillation (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.