University of Science and Technology of China（中国科学技术大学）

Vision-language-action (VLA) models have achieved impressive generalization in robotic manipulation, and recent memory-augmented VLAs have relaxed the Markovian assumption by conditioning on past images or language summaries. Vision-based memory approaches address this by conditioning on sampled past image frames, but they are computationally expensive and fundamentally limited when temporal events are visually ambiguous, e.g., pushing a button multiple times with small movements. We propose FM-VLA, a VLA model with force-based memory, enabling temporal context reasoning for non-Markovian, contact-rich manipulation. We encode force histories into compact force memory tokens with a variational autoencoder (VAE) pretrained with force time series reconstruction. By projecting force latent representations and short state history as additional conditioning tokens to the action expert module, we enable VLAs to leverage accumulated contact event history to guide manipulation. We evaluate FM-VLA on three memory-dependent tasks, including finding a hidden block, pressing a button, and wiping a dish for a specific number of times. Our lightweight force memory achieves over 80% success rate with minimal inference overhead, significantly outperforming baseline approaches. Project page: https://qft-333.github.io/FM-VLA-Page/

Can large language models with substantially different parameter spaces be merged by direct weighted averaging, without training or semantic alignment? Existing heterogeneous fusion methods typically introduce distillation, adapters, learned latent spaces, routing, or feature alignment, leaving open whether a simpler recipe can work for genuinely different billion-parameter checkpoints. We revisit this counterintuitive question through training-free dimensional adaptation followed by ratio-controlled interpolation. In union-style merging, we expand the smaller model into the larger parameter space; in intersection-style merging, we truncate the larger model into the smaller parameter space. Across Qwen-family model pairs and benchmarks covering mathematical reasoning, code generation, language understanding, commonsense reasoning, knowledge, and instruction following, deterministic expansion largely preserves the source model function, and small-ratio interpolation can improve over strong source checkpoints by transferring complementary capabilities. However, near-balanced interpolation often collapses, and task-level results reveal a seesaw effect in which gains on some capabilities coexist with regressions on others. These results show that simple parameter averaging, when paired with lightweight dimensional adaptation and carefully controlled ratios, is a surprisingly strong baseline for heterogeneous LLM merging, suggesting that the limits of direct weighted fusion may also bound what more complex heterogeneous merging methods can achieve at scale.

Identity-preserving text-to-video generation aims to synthesize a video that accurately follows a textual description while maintaining the recognizability of a user-specified subject throughout. The IPVG26 challenge extends this framework from a single holistic prompt to a temporally structured specification. The model additionally receives a sequence of timestamped action captions and must render the subject performing these actions in the specified order. This temporal structure presents a challenge not encountered in previous identity-preserving generation tasks, as the subject must continuously perform a scripted sequence of distinct actions while maintaining a consistent identity. However, end-to-end video generators are prone to appearance drift as motion accumulates and the depicted actions change. We address this challenge with a training-free, three-stage pipeline framework. An action-aware prompt polishment stage first rewrites the inputs into image-generation prompts that specify the terminal state of each action. An identity-preserving generation stage then produces the keyframe sequence by conditioning each frame jointly on the reference identity and its predecessor, thereby decoupling time-invariant appearance from time-varying pose. Finally, an identity-aware inference enhancement stage synthesizes the intermediate segments using multi-reference guidance and identity-driven noise searching, both of which reinforce identity fidelity during sampling. Our method ranked third on the official Track 2 leaderboard, demonstrating competitive performance and strong generality.

Monocular geometry estimation has recently achieved impressive performance across diverse scenes. However, state-of-the-art models still face notable distortion in local 3D structure, especially in fine details, like thin structures and small objects. We attribute this limitation to an architectural mismatch: most current models decode 3D geometry within a 2D parameterization, where feature interactions are governed by image-plane proximity rather than true 3D spatial relationships. This inadvertently mixes features from geometrically distant surfaces, resulting in over-smoothed geometry particularly around thin or elongated structure. In this paper, we propose a fine-detail monocular geometry estimation with Self-Guided Sparse 3D Refinement (SSR) that lifts monocular geometry modeling from 2D image space to 3D space for high-fidelity metric-scale point maps. Our model lifts the coarse point map from a foundation base model onto a sparse voxel shell and refines it via SSR. The SSR employs sparse convolutions that aggregate features based on 3D spatial locality, avoiding feature mixing across depth discontinuities. Extensive experiments on diverse datasets demonstrate that our method significantly outperforms existing approaches in recovering fine detailed 3D geometry across both quantitative metrics and qualitative visualizations.

AI companions are judged not only by single-turn fluency but by whether they sustain emotional continuity: remembering who the companion is, what the user prefers, and how the relationship has felt. We present ZifaMem, a structured memory system that organizes dialogue into session summaries, episodic memories, and a consolidated user model. Against a deployment-honest comparator that supplies the full raw dialogue history, and under a fixed LLM-as-a-judge protocol with route audits, structured memory raises pooled four-backbone emotional-intelligence scores by 11.4% (95% CI 6.3% to 17.1%), and persona grounding improves on all four backbones (Claude +42% relative). Multi-turn affect context wins a +39% net preference over a single-turn snapshot (exploratory), whereas an additional emotion state machine yields no measurable gain on any of five endpoints. Under an identical preregistered protocol, three memory systems (ZifaMem, Mem0, and filtered verbatim retrieval) each improve significantly over raw-history deployment, and ZifaMem and Mem0 are statistically equivalent within +/-5 points on the preregistered primary preference endpoint. The ZifaMem SDK, CLI, and portable Agent Skills are open-sourced at https://github.com/zifacorp/zifamem.

Video multimodal large language models (MLLMs) can describe what happens in a video, but rarely identify when the supporting evidence occurs. We study generalist video temporal grounding, in which one model predicts a variable-cardinality set of evidence intervals across video lengths, domains, query forms, and viewpoints. Existing training strategies are misaligned with this set-valued task: long-video labels often rely on brittle one-pass annotation, while reinforcement-learning rewards either fail to distinguish non-overlapping predictions or require fragile segment matching. TimeLens2 treats temporal evidence as an interval set throughout supervision and optimization. TimeLens2-93K constructs reliable multi-span supervision through caption-derived proposals, independent localization, cross-agent consensus, semantic verification, and boundary refinement. Our temporal Wasserstein reward computes exact one-dimensional \(W_1\) between uniform distributions over merged interval supports, providing dense, matching-free feedback under unequal cardinalities and equivalent fragmentation; temporal IoU complements it with precise-overlap feedback. Across seven benchmarks, TimeLens2-2B outperforms all size-matched baselines on every benchmark, while the 4B and 8B variants achieve state-of-the-art performance, surpassing open-source models with up to 397B parameters. The 2B, 4B, and 8B variants improve over their Qwen3-VL backbones by 14.2, 13.0, and 18.1 mIoU points, respectively.

Deep research agents increasingly operate over the open web, where relevant records coexist with redundant summaries, outdated reports, and misleading documents. Existing evaluations offer limited insight into whether agents preserve sound evidential standards when an ordinary-looking false document is deliberately seeded into a searchable environment and offers a direct shortcut to a conflicting answer. We introduce DRNOISE, a 100-task benchmark for answer recovery under misleading evidence. Each task has a unique gold answer supported by two corroborating indirect record chains; the paired noisy condition adds one plausible document that states a conflicting answer directly. The benchmark spans ten families of evidence operations. Across agents with strong clean-task performance, this single intervention causes 66-88 percentage-point accuracy drops. Trace analyses identify verification inertia as the dominant failure mode: agents often retrieve truthful records but stop before completing and reconciling the evidence chain, instead deferring to the answer-like document. Generic verification prompts reduce but do not close this gap. The setting is especially relevant to open-web deployment, where plausible falsehoods arrive through ordinary-looking pages rather than explicit attacks. Reliable deep research therefore requires more than retrieval and citation; it requires active reconciliation of direct claims with record-level evidence.

Recently, text-to-video (T2V) models have been widely deployed, sparking growing concerns over their robustness against jailbreak attacks. Existing jailbreak methods, mostly adapted from text-to-image attacks, suffer notable drawbacks when applied to T2V systems. They fail to fully leverage temporal consistency, an inherent characteristic of video generation. Besides, these methods demand heavy video query optimization, which is infeasible in practical black-box scenarios. Their adversarial prompt search is also driven by heuristic local signals, lacking principled structured exploration strategies. To tackle these limitations, we propose BSB, a structured, query-efficient jailbreak framework for T2V models. BSB harnesses temporal consistency by encoding harmful intent as the transition between two individually harmless boundary states. Under this paradigm, the attack targets boundary-state pairs whose interpolation tends to produce unsafe intermediate frames during video generation. Directly evaluating all candidate pairs within the video space incurs prohibitive computation cost. Instead, BSB conducts Monte Carlo Tree Search (MCTS) in a cheaper textual proxy space and regularly calibrates search outcomes with sparse video-level evaluations. We conduct comprehensive experiments on mainstream commercial T2V models including Veo 3.1, Sora 2, Seedance and Kling v1. Results show BSB surpasses all existing jailbreak baselines, delivering an average 18.6% relative gain in attack success rate over the strongest competitor across evaluated models. Our findings identify temporal consistency as an understudied yet vital attack surface for T2V models and verify that structured search facilitates effective vulnerability discovery under constrained query budgets.

Human-like private chat requires more than fluent response generation: a system must preserve persona, relationship, memory, bounded knowledge, medium-specific timing, and a coherent multi-turn arc. We present AnthroDial, a closed-loop framework that formulates anthropomorphic dialogue as a joint problem of system architecture, executable evaluation, and diagnostic alignment. It combines (1) a role-conditioned scheduled dialogue runtime with persona and scenario cards, long-term memory, virtual time, and single-draft message decisions; (2) an executable benchmark with an L0 validity gate, five per-turn dimensions, and five dialogue-level dimensions; and (3) a post-training pipeline that filters 16,436 scheduled-decision examples for SFT and applies GRPO with a cognitive-diagnostic, ZPD-aware reward. The reward maintains Kalman-filtered capability estimates for each behavioral dimension, upweights dimensions with larger capability deficits, and uses rollout scores as task-level ZPD matches to focus optimization on learnable weak skills. On a benchmark with 55 personas, 50 scenarios, 50 persona-scenario bindings, and 100 role-conditioned cases per model, we evaluate 16 systems spanning frontier baselines, open models, thinking/no-think variants, and SFT/RL ablations. The strongest non-trained baseline reaches 32.00% strict ACC, while Qwen3.6-27B-SFT+RL reaches 39.00% strict ACC and a 98.5 overall score. In the 9B no-think setting, SFT and RL improve strict ACC from 0.00% to 13.00% and 18.37%. These results show that anthropomorphic dialogue benefits when generation, evaluation, and reward shaping share the same behavioral dimensions.

Test-time scaling empowers Large Reasoning Models (LRMs) to tackle complex tasks via extensive Chain-of-Thought (CoT). However, this often induces the "overthinking" paradox, where redundant reasoning increases computational overhead without guaranteeing accuracy. Existing test-time efficiency optimization methods primarily fall into two categories: information-theoretic approaches, which are prone to "deceptive convergence" where low uncertainty masks hallucinations, and latent representation analyses, which are often post-hoc, lacking the real-time sensitivity for dynamic reasoning. To bridge this gap, we first posit the Phase-Momentum Alignment Hypothesis, asserting that reasoning correctness hinges on the temporal synchronization between geometric momentum and uncertainty resolution. We then theoretically formulate the Cognitive-Energy Model to characterize these dynamics through two orthogonal dimensions: Geometric Cognitive Effort, quantified by latent velocity and tortuosity, and Entropic Cognitive Uncertainty. To operationalize this, we introduce PUMA (Phase-Uncertainty Momentum Alignment), a training-free framework employing a tiered diagnostic architecture. By coupling lightweight phase monitoring with event-triggered geometric analysis, PUMA effectively distinguishes active exploration from passive stagnation, enabling precise interventions through adaptive truncation or corrective measures. Extensive experiments on LRMs spanning 1.5B to 32B demonstrate that PUMA consistently outperforms state-of-the-art baselines across diverse benchmarks, achieving a superior accuracy-efficiency trade-off and robust cross-domain generalization.

Surgical video phase recognition is a fundamental task in computer-assisted intervention, supporting workflow understanding, intraoperative guidance, and surgical quality assessment. Although recent visual-temporal models have achieved promising progress, accurate and temporally coherent phase recognition remains challenging due to local visual ambiguity, transient prediction noise, and insufficient use of procedural semantics. To address these challenges, we propose HTT-Net, a Hierarchical Text-guided Transition modeling Network for surgical video phase recognition. The key idea is to introduce structured surgical semantic knowledge into phase-aware segment construction and semantic refinement. Specifically, we construct a hierarchical surgical semantic memory with intra-phase descriptions, inter-phase transition descriptions, and fine-grained semantic units. Based on this memory, the proposed Transition-Aware Segment Construction (TAS-Con) organizes frame-level evidence into coherent segment representations and handles boundary clips with inter-phase transition descriptions. Furthermore, we introduce Transition-Aware Segment Calibration (TAS-Calib), which calibrates phase-aware segment representations through hierarchical surgical semantics and improves discrimination under visual ambiguity without dense frame-level vision-language fusion. Experiments on Cholec80 and LCRS-100 demonstrate the effectiveness of HTT-Net for robust surgical video phase recognition.

Existing memory systems for long-horizon LLM agents often retrieve prior traces as passive context rather than converting them into executable capabilities. In this paper, we propose MSCE, a training-free Memory--Skill Co-Evolution framework that organizes agent experience into grounded step traces, reusable procedural policies, and declarative environmental cognition. MSCE crystallizes evidence-backed L2 policies with positive estimated gain into callable skills that retain evidence links, applicability boundaries, decision guidance, verification rules, and reliability estimates. It further introduces reflection-weighted value backfilling, which propagates sparse terminal feedback through dense local self-reflections to produce evidence-calibrated trace values for governing memory and skill evolution. Experiments on EvoAgentBench and LoCoMo demonstrate that MSCE significantly outperforms state-of-the-art skill-augmented and memory-driven agent baselines, exhibiting strong cross-domain transferability and lifelong-evolution capabilities.

Medical Vision-Language Models (Med-VLMs) require reliable reasoning from fine-grained visual evidence, yet existing models can produce plausible clinical answers by relying on language priors or medical templates rather than truly attending to diagnosis-critical regions. On-Policy Distillation (OPD) offers dense token-level supervision on student-generated trajectories and provides a privacy-compatible means of capability transfer without requiring the redistribution of raw patient data. However, standard OPD uniformly distills all tokens, causing sparse evidence-dependent tokens to be diluted by abundant clinical narrative tokens. Inspired by the success of OPD in the large language model community, we propose \textbf{Med-OPD}, to our knowledge the first unified post-training framework that integrates on-policy distillation with medical evidence-aware supervision for Med-VLMs. We introduce \textbf{Medical Evidence Advantage} (MEA), a teacher-grounded counterfactual signal that uses an answer-aware hint to focus teacher scoring on evidence supporting the target diagnosis, and measures each token's dependence on medical visual evidence by comparing teacher likelihoods under the original and evidence-degraded imaging modalities. Based on MEA, Med-OPD redistributes the distillation signal at both the token and trajectory levels, emphasizing diagnosis-critical tokens and evidence-reliant rollouts. Experiments on OmniMedVQA subsets show that Med-OPD consistently outperforms SFT and standard OPD across CT, MRI, Disease Diagnosis, and Lesion Grading. These results demonstrate that evidence-aware distillation can better strengthen medical VLMs' reliance on key visual evidence and improve reliable multimodal medical reasoning. The source code and data is publicly available at: https://github.com/yunhang8658/MedOPD.git

Category-level object pose estimation is a crucial yet challenging task in both academia and industry, and has achieved remarkable success by leveraging keypoint-based correspondence paradigms. However, most existing methods increasingly rely on stronger feature learning while overlooking whether the established correspondences are geometrically stable across diverse perturbations. This often results in fragile pose recovery under intra-class shape variations and occlusions. To tackle this challenge, we develop a novel Triangle-Invariant Geometric Consistency Learning for Category-Level Object Pose Estimation (TriCons-Pose) to anchor stable keypoints and aggregate pose-invariant cues, yielding reliable canonical mapping and accurate pose estimation. Specifically, a Structure-Consistent Keypoint Detector (SCKD) is designed to identify robust keypoints by enforcing cross-view structural consistency via normalized pairwise distance matching. Moreover, we propose a Pose-Invariant Geometric Aggregator (PIGA) to augment keypoint representations by injecting triangle-based pose-invariant descriptors into a local-to-global attention mechanism. The proposed framework is optimized using standard objective functions while incorporating an additional geometry consistency loss. Extensive experiments on REAL275, CAMERA25, and HouseCat6D datasets demonstrate the effectiveness of the proposed approach.

Vision-Language Models (VLMs) are costly at inference time because they must process long sequences of visual tokens. Existing token pruning methods often degrade under high compression by blindly discarding information, breaking spatial structure or collapsing diversity. We propose SpecFlow, a training-free framework that shifts the paradigm from destructive pruning to conservative condensation, strictly enforcing spatial coverage and statistical conservation to ensure stability. Treating visual tokens as nodes in a $k$NN graph, SpecFlow (i) computes a stable importance field via spectral heat flow to preserve structural coherence, (ii) allocates budgets via adaptive spatial partitioning to guarantee coverage, and (iii) aggregates discarded information into coreset sinks to maintain statistical conservation. The method is plug-and-play, requires no fine-tuning, and is compatible with FlashAttention. Experiments confirm that our SpecFlow outperforms SOTA methods across tasks, VLM architectures, and pruning ratios. Notably, LLaVA-1.5 with SpecFlow retains 95.6% of original performance despite pruning 88.9% of visual tokens, offering an exceptional efficiency-accuracy balance. Code is available at https://github.com/Lzy-dot/SpecFlow

We find that current emotion-oriented Omni-MLLMs still lack reliable omni-modal perception: they (i) underutilize multimodal cues in their reasoning trajectories and (ii) exhibit unfaithful behavior, often hallucinating modality-specific statements from other modalities. Building on these insights, we propose OPPO (Omni-Perception Policy Optimization), a reinforcement learning framework that explicitly optimizes multimodal perception. First, an Omni-Perception Reward decomposes ground-truth reasoning into fine-grained visual, acoustic, and emotion cues and rewards trajectories that semantically recover these cues. Second, an Omni-Perception Loss compares the policy under full and unimodally masked inputs, applying a KL penalty only to modality-specific evidence tokens to suppress cross-modal hallucination. We further introduce MEP-Bench, a diagnostic benchmark that quantifies utilization and faithfulness. Experiments show that OPPO achieves state-of-the-art performance on MER-UniBench and MME-Emotion, while substantially improving utilization and faithfulness scores on MEP-Bench, highlighting the importance of sufficient and faithful omni perception for multimodal emotion reasoning.

In this report, we present Hy-Embodied-0.5-VLA, abbreviated as HyVLA-0.5, an end-to-end system that spans the full robot learning stack: data collection, model design, continued pre-training and supervised fine-tuning, RL post-training, and real-world deployment. Each component serves a distinct role in this stack.

We present an interpretable machine learning pipeline to decompose cross-sectional equity return predictability into auditable factor contributions. We apply an XGBoost model with TreeSHAP attribution and conduct stress testing on 3,632 Chinese A-share stocks from 2009 until 2019. On prediction, using 60-month rolling windows over 55 months of out-of-sample data, XGBoost obtains a mean AUC of 0.547 (rank IC = 0.119) and +2.38%/month (Newey-West t = 5.94; annualized Sharpe 2.23) long-short spread for the top vs bottom quintiles. This alpha is persistent after adjusting for the Carhart four-factor model (+2.31%/month; t = 7.48). On interpretation, SHAP decomposition indicates that behavioral signals (turnover and momentum) account for 58.2% of predictive attribution compared to 10.7% for valuation ratios, on average, across 50 industry groups. Ablation analysis serves to cross-validate this ranking and provides evidence that SHAP and ablation diverge in a manner that highlights feature substitutability structure that is largely invisible to either method used in isolation.

Question answering (QA) systems have achieved notable progress with the advent of large language models (LLMs). However, they still face challenges in accurately extracting and generating precise answers from given contexts, particularly when dealing with complex or ambiguous queries. Existing approaches often struggle with contextual understanding, answer consistency, and generalization across diverse domains. In this work, we propose a question answering system based on large language models, where the input consists of a textual context and a corresponding question, and the output is a concise and accurate answer. The motivation behind this research lies in addressing the limitations of current QA systems, particularly their tendency to produce irrelevant or imprecise responses despite having access to the correct context. Our methodology involves fine-tuning a pre-trained LLM on a benchmark QA dataset to improve its contextual comprehension and answer extraction capabilities. Specifically, we utilize the Stanford Question Answering Dataset (SQuAD1.1), which provides high-quality context-question-answer triplets for supervised training and evaluation. Experimental results show that the fine-tuned Roberta-base model achieves the highest performance, attaining a ROUGE-L score of 86.84%, a BLEU score of 28.24%, and a BERTScore of 95.38%. These results indicate strong accuracy and answer relevance, demonstrating the effectiveness of the proposed approach for context-based question answering tasks. Furthermore, the findings confirm that targeted fine-tuning substantially improves the reliability and precision of QA systems.

Vision-Language-Action (VLA) models offer a promising paradigm for generalist robotic policies, yet their adaptation is hindered by data inefficiency and poor generalization. We argue that these bottlenecks stem from the prevailing Direct Instruction-to-Control Mapping, which forces models to memorize monolithic trajectories rather than reusable motion patterns, i.e., primitives. We propose PrimitiveVLA, a framework that shifts this paradigm toward a Primitive-Centric Disassemble & Assemble paradigm. Supported by a shared Multimodal Canonical Representation (MCR), PrimitiveVLA unifies two phases: (1) Fine-tuning-phase Disassembly, which uses an automated pipeline to disassemble demonstrations into reusable primitives; and (2) Inference-phase Assembly, which employs a VLM-based planner and an LLM-generated switch module for robust closed-loop execution. By disassembling tasks into reusable primitives, PrimitiveVLA enables VLA models to learn invariant motion patterns instead of task-specific trajectories. Extensive experiments show that our framework improves data efficiency and achieves superior zero-shot generalization across unseen and long-horizon tasks.

Loss spikes remain a persistent obstacle in large-scale language model pretraining. While previous research has attempted to identify the root cause of loss spikes by investigating individual factors, we observe that, in practice, such spikes are typically triggered by the confluence of heterogeneous factors. Empirically, loss spikes may arise from a combination of data outliers, hardware or transient computational faults, numerical precision issues, and hyperparameter settings. Regardless of the underlying cause, these spikes manifest as unstable optimizer updates, as abnormal gradients contaminate both first- and second-moment states. In this paper, we propose a principled gradient-centric remedy: AdaGC, an adaptive per-tensor gradient clipping scheme that mitigates such contamination by bounding gradient norms relative to a tensor-wise exponential moving average of their historical clipped values. AdaGC is optimizer-agnostic, introduces negligible memory overhead, and reduces communication costs compared to GlobalGC, particularly in hybrid-parallel distributed training. Experiments on Llama-2 7B, Mixtral 8x1B, and ERNIE 10B-A1.4B demonstrate that AdaGC robustly eliminates training instabilities, consistently reducing spike scores to zero for all models and improving downstream accuracy over GlobalGC by 1.32%, 1.27%, and 2.48%, respectively. Furthermore, AdaGC seamlessly integrates with optimizers such as Muon and Lion, consistently yielding higher average accuracy and zero spike scores. The code is available at https://github.com/PaddlePaddle/PaddleFleet (see Research/AdaGC).

Glass surfaces are ubiquitous in daily life, typically appearing colorless, transparent, and lacking distinctive features. These characteristics make glass surface detection a challenging computer vision task. Existing glass surface detection methods always rely on boundary cues (e.g., window and door frames) or reflection cues to locate glass surfaces, but they fail to fully exploit the intrinsic properties of the glass itself for accurate localization. We observed that in most real-world scenes, the illumination intensity in front of the glass surface differs from that behind it, which results in variations in the reflections visible on the glass surface. Specifically, when standing on the brighter side of the glass and applying a flash towards the darker side, existing reflections on the glass surface tend to disappear. Conversely, while standing on the darker side and applying a flash towards the brighter side, distinct reflections will appear on the glass surface. Based on this phenomenon, we propose NFGlassNet, a novel method for glass surface detection that leverages the reflection dynamics present in flash/no-flash imagery. Specifically, we propose a Reflection Contrast Mining Module (RCMM) for extracting reflections, and a Reflection Guided Attention Module (RGAM) for fusing features from reflection and glass surface for accurate glass surface detection. For learning our network, we also construct a dataset consisting of 3.3K no-flash and flash image pairs captured from various scenes with corresponding ground truth annotations. Extensive experiments demonstrate that our method outperforms the state-of-the-art methods. Our code, model, and dataset will be available upon acceptance of the manuscript.

Vision-Language Models (VLMs) show great promise for autonomous driving, but their suitability for safety-critical scenarios is largely unexplored, raising safety concerns. This issue arises from the lack of comprehensive benchmarks that assess both external environmental risks and in-cabin driving behavior safety simultaneously. To bridge this critical gap, we introduce DSBench, the first comprehensive Driving Safety Benchmark designed to assess a VLM's awareness of various safety risks in a unified manner. DSBench encompasses two major categories: external environmental risks and in-cabin driving behavior safety, divided into 10 key categories and a total of 28 sub-categories. This comprehensive evaluation covers a wide range of scenarios, ensuring a thorough assessment of VLMs' performance in safety-critical contexts. Extensive evaluations across various mainstream open-source and closed-source VLMs reveal significant performance degradation under complex safety-critical situations, highlighting urgent safety concerns. To address this, we constructed a large dataset of 98K instances focused on in-cabin and external safety scenarios, showing that fine-tuning on this dataset significantly enhances the safety performance of existing VLMs and paves the way for advancing autonomous driving technology. The benchmark toolkit, code, and model checkpoints will be publicly accessible.

Balancing training accuracy and adversarial robustness has beeen a challenge since the birth of deep learning. Here, we introduce a geometry-aware deep learning framework that leverages layer-wise local training to sculpt the internal representations of deep neural networks. This framework promotes intra-class compactness and inter-class separation in feature space, leading to manifold smoothness and adversarial robustness against white or black box attacks. The performance can be explained by \blue{data-dependent statistical mechanics of integrating out the network parameters}, \blue{supplemented by a phenomenological model} with Hebbian coupling between elements of the hidden representation. Based on the current geometry-aware learning framework, the deep network can assimilate new information into existing knowledge structures while reducing representation interference.

Retrieval-Augmented Generation (RAG) has gained significant attention in recent years for its potential to enhance natural language understanding and generation by combining large-scale retrieval systems with generative models. RAG leverages external knowledge sources, such as documents, databases, or structured data, to improve model performance and generate more accurate and contextually relevant outputs. This survey aims to provide a comprehensive overview of RAG by examining its fundamental components, including retrieval mechanisms, generation processes, and the integration between the two. We discuss the key characteristics of RAG, such as its ability to augment generative models with dynamic external knowledge, and the challenges associated with aligning retrieved information with generative objectives. We also present a taxonomy that categorizes RAG methods, ranging from basic retrieval-augmented approaches to more advanced models incorporating multimodal data and reasoning capabilities. Additionally, we review the evaluation benchmarks and datasets commonly used to assess RAG systems, along with a detailed exploration of its applications in fields such as question answering, summarization, and information retrieval. Finally, we highlight emerging research directions and opportunities for improving RAG systems, such as enhanced retrieval efficiency, model interpretability, and domain-specific adaptations. This paper concludes by outlining the prospects for RAG in addressing real-world challenges and its potential to drive further advancements in natural language processing.

Multimodal Knowledge Graph Completion (MKGC) requires inferring missing entities from structural, textual, and visual cues. Existing diffusion-based MKGC methods usually denoise directly on raw multimodal features. Such a design forces the denoiser to simultaneously perform relation-dependent cue selection, cross-modal semantic alignment, and structure-aware entity generation, which introduces noisy and semantically inconsistent conditions for diffusion and consequently leads to suboptimal completion performance. To address this limitation, we propose MGDT: MLLM-Guided Diffusion Transformer with Relation-Adaptive Mixture-of-Experts (MGDT), a novel MKGC framework built on an align-then-diffuse paradigm. MGDT first employs a Relation-Adaptive Semantic Routing Mixture-of-Experts (RASR-MoE) module to select relation-relevant multimodal semantic transformation paths and suppress irrelevant modality interference. MGDT then uses a frozen Multimodal Large Language Model (MLLM) as a semantic anchor to align the routed multimodal representations into a unified latent space and reduce cross-modal semantic heterogeneity. Finally, a Knowledge Graph Diffusion Transformer (KGDT) performs graph-conditioned denoising generation in the aligned space to produce the missing entity representation. Experiments on three benchmark datasets show that MGDT consistently outperforms strong baselines.

Large language model agents increasingly rely on skills: reusable procedural documents encoding workflows, tool use, implementation patterns, validation checks, and domain rules. Skill rewriting is often treated as prompt compression, but shorter skills can make agents more expensive by removing sparse operational anchors that prevent exploration, debugging, and recovery. We study skill rewriting through this economic lens. Our controlled framework profiles skill structure, rewrites skills using information-preservation strategies, and evaluates the rewrites under fixed task instructions, environments, and verifiers. Experiments on SkillsBench reveal distinct quality--cost trade-offs across strategies: API/code anchoring, workflow guarding, and rule/formula anchoring benefit different task families, with no universally dominant template. In the main held-out evaluation, the learned policy reduces total cost by 7.0% and downstream agent-token cost by 6.0%; in frozen cross-model transfer, the corresponding reductions average 14.7% and 13.7%, while verifier quality is preserved. These results position skill design as cost-aware operational knowledge engineering rather than prompt compression. Resources: https://github.com/1Reminding/Skill_EE.

Long-form video editing over heterogeneous footage requires agents to coordinate source selection, multimodal analysis, timeline construction, narration and subtitle alignment, rendering, and revision while exposing intermediate state for inspection and repair. We present Crayotter, an open-source multimodal multi-agent demo system for prompt-driven long-form video editing. Crayotter organizes production around coverage-aware material preparation, artifact-grounded editing research, and tool-grounded timeline execution. Across these stages, retrieval reports, video analyses, editing blueprints, scheduler events, tool calls, intermediate renders, and final exports are treated as first-class artifacts rather than hidden transient state. The workbench supports local assets, agent-assisted retrieval, progress monitoring, artifact preview, failure diagnosis, interrupted-job resumption, and resource-aware asynchronous execution for long-running workflows. In a 23-theme evaluation, Crayotter achieves the highest human overall score (3.40/5) among the compared systems, with its largest margins in theme alignment, narrative coherence, and editing smoothness. These results show that long-horizon video editing agents can be made traceable, inspectable, and practically controllable through observable production artifacts. Code, traces, and examples are publicly available at https://github.com/idwts/Crayotter.

Group Relative Policy Optimization (GRPO) enables stable and preference-oriented updates via group-wise comparisons for post-training video generation. However, GRPO directly optimizes reward-induced advantages. Under sustained optimization, the reward score can lose fidelity as a proxy for true video quality, consistent with the phenomenon described by Goodhart's Law. This leads to two recurring issues: (i) shortcut-driven optimization under composite objectives and (ii) reward saturation within prompt groups. To address these issues, we introduce TaRoS, a Target-Robust Reward Signaling framework for Video generation GRPO. TaRoS leverages component level performance assessment together with intra-group sparsity to organize multi-aspect rewards towards optimization objectives. In addition, it adaptively downweights components that exhibit saturation, thereby preserving effective optimization directions and mitigating redundancy. This maintains meaningful optimization directions and preserves within-group ranking separation, thereby preventing reward hacking and leading to more reliable policy updates. Extensive experiments show consistent improvements in visual fidelity, motion coherence, and text-video alignment over strong baselines.

Political discourse has increasingly moved to short-video platforms, yet computational analysis of such content remains constrained by the scarcity of datasets that jointly preserve audiovisual information and hierarchical conversations. Here we present TikStance, a multimodal and context-aware dataset comprising 161 videos and 13,876 comments from TikTok, designed for stance detection in political discussions. The dataset covers three major political figures in the 2024 U.S. election cycle--Donald Trump, Joe Biden, and Kamala Harris--with content collected between September 2023 and January 2025. Each discussion unit links a host video and its metadata to a parent-linked comment tree, enabling stance analysis within both audiovisual and conversational context. Each item was independently labeled by three annotators using a three-class scheme (Favor, Against, None) for video-to-target and comment-to-target stance; items with disagreement were re-annotated, and the final Krippendorff's \(α\) reached 0.743, 0.723, and 0.722 for the Trump, Biden, and Harris subsets, respectively. Descriptive analysis further reveals target-dependent differences in stance distributions and conversational depth, with nested replies accounting for 23.3\% of all comments. By combining multi-target coverage, hierarchical conversations, and reliable multi-level human annotations, TikStance supports research in multimodal stance detection, political communication, computational social science, and context-aware natural language processing.