Y Shi

Ruhan Wang, Yucheng Shi, Zongxia Li et al.

The capability of a modern AI agent depends not only on its foundation model but also on its harness, which constructs prompts, manages state, invokes tools, and coordinates execution. As models, APIs, environments, and requirements evolve, the harness must be continually modified. Before such a change can be made, a developer or coding agent must identify all code locations that implement the target behavior. This is difficult because production harnesses are large, tightly coupled, and behaviorally distributed, while modification requests describe what the system should do and repositories are organized by files and modules. Code search, repository indexing, and long-context processing ease inspection, but still leave this behavior-to-code mapping to be recovered by hand. Behavior localization is therefore a central bottleneck in harness evolution. We introduce the Harness Handbook, a behavior-centric representation synthesized automatically from a harness codebase via static analysis and LLM-assisted structuring, linking each behavior to its corresponding source. We also introduce Behavior-Guided Progressive Disclosure (BGPD), which guides agents from high-level behaviors to relevant implementation details and verifies candidate locations against the current source. On diverse modification requests from two open-source harnesses, Handbook-Assisted planning improves behavior localization and edit-plan quality while using fewer planner tokens, with the largest gains on scattered sites, rarely executed paths, and cross-module interactions. Evolving complex agentic systems thus depends not only on generating edits, but also on determining where those edits should be made.

Xinyu Liu, Shihao Li, Weihong Lin et al.

Recent diffusion-based video generation models have made significant progress in multi-reference image-conditioned video editing. However, existing methods still struggle to coordinate information from multiple visual sources accurately. We identify a critical deficiency in existing approaches. Existing editing instructions lack explicit reference relationships, and most multimodal large language models (MLLMs) cannot generate them reliably. To address this problem, we propose ReBind, a systematic framework that introduces semantic instructions with embedded reference tokens as the intermediate representation for multi-reference image-conditioned video editing. Our key insight is embedding reference tokens at semantic positions to eliminate ambiguity and establish precise bindings between visual attributes and their sources. We develop ReBind-Instruct, a specialized MLLM that learns to establish explicit bindings between visual attributes and their reference sources through a two-stage progressive scheme for precise reference relationships. We further develop ReBind-Edit, which enables lightweight adaptation of text-to-video models to coordinate multiple references by binding visual attributes to their designated sources. Extensive experiments demonstrate that ReBind substantially outperforms general-purpose MLLMs in instruction quality and achieves state-of-the-art performance among open-source methods on reference image conditioned video editing. Our project webpage: https://rebind-mrv2v.github.io/.

Yanqiao Zhu, Jingru Gan, Xiaoqi Sun et al.

Multi-step retrosynthesis planning seeks to decompose a target molecule into commercially available building blocks through a sequence of feasible reactions. The vast combinatorial search space makes this task challenging even for expert chemists. Traditional methods combine tree search with offline-trained value networks that score candidates in isolation, without reasoning about complete multi-step routes. Recent work leverages Large Language Models (LLMs) for this task, but relies on simple interfaces that limit exploration of the full search space. We introduce RetroAgent, an LLM agent that bridges symbolic search and neural reasoning through a harness with structured memory. Through memory and chemistry tools, the agent observes the full search state, including explored routes, available alternatives, and properties of intermediates, enabling informed decisions grounded in both global progress and domain knowledge. Experiments on in-distribution and out-of-distribution benchmarks demonstrate that RetroAgent delivers strong performance and generalization.

Yuqi Tang, Tengfei Liu, Yizheng Lai et al.

Video generation increasingly relies on keyframe-based workflows, where creators specify a sequence of reference images to guide generation. Although recent models support multi-keyframe conditioning, it remains unclear whether they can faithfully reproduce the prescribed keyframes while maintaining overall video quality. We present KeyFrame-Compass, the first comprehensive benchmark for evaluating keyframe-conditioned video generation. The benchmark contains 386 carefully curated samples spanning three application domains, two video structures, two prompt granularities, two conditioning formats, and four keyframe densities, enabling controlled analysis under diverse generation settings. We further introduce an automated evaluation framework that jointly measures keyframe execution and overall video quality. Specifically, we decompose keyframe execution into six complementary metrics covering presence, fidelity, temporal ordering, localization, persistence, and uniqueness, while assessing overall video quality through evidence-grounded MLLM judgments augmented with specialized perception models. Experiments on nine representative video generation systems reveal several fundamental limitations. Current models exhibit a clear trade-off between faithful keyframe execution and natural video synthesis. Their performance further degrades as keyframe constraints become denser and most open-source models also fail to interpret storyboard-grid inputs as temporally ordered keyframe sequences.

Junyao Yang, Yucheng Shi, Zongxia Li et al.

Asynchronous reinforcement learning improves throughput by decoupling rollout generation from optimization, but staleness is an inevitable byproduct compounded by policy lag, engine delays, and mixture-of-experts routing. From a trust-region perspective, this mismatch is critical: training-inference divergence governs approximation error in finite-horizon bounds, whereas PPO clipping only gates sampled outward updates, acting as a sampled surrogate rather than a full-policy constraint. As a result, high-staleness updates remain weakly controlled in the asynchronous regime where stale rollouts matter most. We introduce the Staleness-Adaptive Trust Region (SAT), which uses the detached sampled log-ratio as a practical staleness proxy, identifies high-mismatch tails within each batch via staleness-based kernel scaling, and contracts only the sign-selected endpoint of the nominal PPO interval. This preserves baseline behavior on ordinary tokens while enforcing more conservative updates on newly intercepted outward bands. We prove local interval containment and pointwise pessimism relative to PPO, showing how the adaptive rule reshapes update geometry under heterogeneous staleness. We evaluate SAT in a decoupled asynchronous RL setup built on Qwen3-30B-A3B-Base, using SGLang as the inference engine and Megatron for training. In this setting, SAT-GSPO w/ R3 achieves the best observed AIME24 avg@8, reaching 35.83 at lag 1 and 34.79 at lag 8, while SAT-GSPO reaches 34.17 at lag 1. Adaptive clipping and routing replay act as complementary stabilizers targeting mismatch tails and routing inconsistency, respectively. Overall, aligning clip intervals with staleness heterogeneity effectively stabilizes asynchronous RL.

Yue Shi, Liangxiu Han, Xin Zhang et al.

Machine olfaction for gas unmixing is an underconstrained inverse problem in which gas compositions must be inferred from low-dimensional, delayed, and entangled sensor responses produced by interacting chemical transport, surface adsorption, and sensor transduction. One of the key obstacles is physics closure misspecification, where a neural network is designed to fit sensor traces rather than infer a physically closed olfactory process. In this work, we formulate gas unmixing as a multi-physics-constrained inverse problem governed by Maxwell--Stefan multicomponent transport PDEs, competitive adsorption ODEs, and nonlinear sensor transduction ODEs. Directly solving such a high-dimensional coupled system is computationally expensive and often numerically unstable. To this end, we propose UnMixNet, a physics-closed graph neural solver that embeds this multi-physics forward process into end-to-end gas unmixing. UnMixNet discretizes Maxwell--Stefan cross-diffusion on spatial graphs and formulates the multicomponent flux on each edge. This design enables local, differentiable, and flux-conservative inference for multicomponent cross-diffusion. Evaluations on SmellNet show improved single-odor recognition, seen-mixture unmixing, and unseen-mixture generalization. In addition, an external validation on UCI Dynamic Gas Mixtures shows that the inferred concentration process agrees with ground truth concentration set points under dynamic transitions. Process-consistency diagnostics further show that the proposed model learns transferable dynamic physical fingerprints that better satisfies transport, conservation, adsorption, and readout closure.

Xinhao Li, Yuhan Zhu, Xiangyu Zeng et al.

Recent advances in video understanding have spanned motion, long video, and streaming interaction, driving this field toward real-world applications. Despite this progress, current open-source models remain limited in several ways. They often struggle to generalize across diverse video types, making them effective only in specific domains. High computational demands further restrict their efficiency and scalability. Moreover, most models are only partially open, with key components such as training code, strategy, or datasets unavailable, which hinders reproducibility and slows community-driven development. To address these issues, we introduce VideoChat3, a fully open, efficient, and generalist video-centric MLLM. VideoChat3 advances video understanding through two complementary designs. For efficiency, we introduce Inflated 3D Vision Transformer (I3D-ViT) and Adaptive Frame Resolution for Streaming Video Perception, which enables efficient spatiotemporal representation and reduces the cost of processing video inputs during training and inference. For effectiveness, we develop a scalable video data synthesis pipeline that curates three diverse, high-quality training datasets: VideoChat3-Academic2M, VideoChat3-LV116K, and VideoChat3-OL617K, covering general, long-form, and streaming video scenarios, improving the model's generalization across domains. By integrating these designs, VideoChat3 achieves a rare balance of broad generalization and computational efficiency. Experiments across general, long-form, and streaming benchmarks demonstrate that VideoChat3 surpasses prior open-source models with equal or larger parameter counts with only 4B parameters and higher efficiency.

Wenhao Zeng, Yuling Shi, Xiaodong Gu et al.

Program verifiers play a central role in training coding agents, including selecting trajectories for supervised fine-tuning (SFT) and providing rewards for reinforcement learning (RL). Standard execution-based verification requires running unit tests inside per-repository environments such as Docker images, incurring substantial environment setup costs. We propose Dockerless, an environment-free agentic patch verifier that evaluates generated code patches without executing them. Rather than simply matching candidate patches to references, Dockerless judges patch correctness using evidence gathered through agentic repository exploration. On a verifier evaluation benchmark, Dockerless outperforms the strongest open-source verifier by 14.3 AUC points. Using Dockerless as both the SFT trajectory filter and the RL reward enables a fully environment-free post-training pipeline. The resulting model reaches 62.0%, 50.0%, and 35.2% resolve rate on SWE-bench Verified, Multilingual, and Pro, respectively. It surpasses the Qwen3.5-9B baseline by 2.4, 8.7, and 2.9 points, matching environment-based post-training.

Yukun Shi, Minglun Gong

Dynamic-scene reconstruction is almost always evaluated inside the observed time window, yet deployment settings such as AR overlays, robot interaction, and anticipatory planning need the future surface: the geometry at times beyond those captured. No standard benchmark measures this. We introduce FutureSurf, a controlled diagnostic benchmark and dataset for future-time surface reconstruction that trades scene diversity for exact future ground truth and falsification controls. A method trains on the observed first 75% of a sequence; we score its extracted per-frame surface on the held-out future by Chamfer distance, reporting absolute future CD as the primary score and the future/observed gap as a diagnostic. The dataset contains eight analytically defined controlled motions, including three falsification controls, with exact per-frame ground-truth meshes. We also provide a ground-truth-side recoverability oracle. The release includes split files, scoring code, a benchmark card, and Croissant metadata. On the controlled motions, the DG-Mesh backbone leaves a 2.7-4.1$\times$ gap even for futures predictable in principle (four of five recoverable from observed motion by a fixed rule), while the falsification controls behave as designed (the surface-invariant motion shows no gap). Beyond the contributed dataset, the gap persists across six animated DG-Mesh asset scenes and a second backbone, Deformable-3DGS (2.0-6.6$\times$; both share a deformation-MLP temporal model). The benchmark also shows that future rendering quality and future-surface accuracy are statistically decoupled, so the novel-view-synthesis metrics the field reports do not track future geometry. The future error is structured, concentrating where the surface moves. The dataset, evaluation toolkit, and scoring code are available on Hugging Face and GitHub (https://github.com/Ricky-S/futuresurf).

Xiaohan Zhang, Yuqing Wen, Junlin Chen et al.

Multi-reference-to-audio-video (MR2AV) generation aims to generate coherent audio-video content conditioned on multiple references and textual instructions. Existing benchmarks mainly focus on text-driven generation, single-reference subject preservation, or isolated audio-video alignment, leaving the emerging MR2AV setting largely unexplored. Compared with these settings, MR2AV requires models to jointly reason over multiple references while generating synchronized visual and audio content. Models must not only preserve each reference faithfully but also correctly bind and compose multiple referenced entities into coherent audio-visual events. To address this gap, we introduce MultiRef-Compass, a unified benchmark for MR2AV generation. It comprises $350$ carefully curated samples constructed through a scalable and controllable asset-composition pipeline, covering multi-view subject preservation, multi-entity binding, and human-object-scene composition. To provide interpretable assessment, MultiRef-Compass defines an evaluation protocol with four dimensions: Basic Quality, Reference Consistency, Audio-Visual Consistency, and Instruction Following, using 14 sub-metrics. MultiRef-Compass integrates automatic metrics with a rejudging-enhanced MLLM-as-a-Judge framework, enabling scalable and auditable evaluation of both perceptual fidelity and reference-conditioned composition. Extensive experiments on eight representative MR2AV systems reveal substantial room for improvement across multiple evaluation dimensions, underscoring the need for a comprehensive benchmark and positioning MultiRef-Compass as a foundation for future MR2AV research.

Jun Zhuang, Mohammad Al Hasan, Yiyu Shi et al.

Noisy-label learning in small-scale medical image classification is challenging and hinders the superiority of deep neural networks. Recent studies suggest that quantum neural networks (QNNs) have shown potential in limited-data regimes, yet their use for noisy-label learning remains under-explored. A key obstacle is QNNs' intrinsic "natural smoothness", which may regularize training but also obscure high-confidence samples needed for noise-transition estimation. We propose Supermartingale-based Label Transition (SLT), an anchor-free loss correction framework for robust QNN-based medical image classification under noisy labels. SLT models entropy reduction in predictive distributions as a supermartingale and uses its monotonic behavior to identify stable transition-matrix refinement steps. This enables dynamic transition updates while reducing noise-driven oscillations during QNN training. We further provide a convergence analysis showing that the proposed transition-refinement process reaches a steady state. Experiments on multiple public small-scale medical image datasets demonstrate that SLT consistently improves QNN-based classification and stably outperforms classic noise-label learning baselines under synthetic and real-world label noise.

Mohaimin Al Barat, Hexuan Yu, Shaoyu Li et al.

Dynamic Spectrum Sharing (DSS) is a cornerstone of next-generation wireless systems, yet existing solutions such as Spectrum Access Systems (SAS) rely on centralized administrators that expose sensitive operational metadata and lack cryptographic transaction accountability. Though SAS administrators, such as Google, have introduced pay-as-you-go pricing models, these approaches still face significant privacy and accountability challenges as DSS evolves toward a more open and large-scale spectrum marketplace. We present SpexPay, a privacy-preserving and auditable pay-as-you-go spectrum usage framework that enforces fine-grained, usage-linked payments without revealing user identities. Spexpay integrates BBS+ verifiable credentials, unlinkable session pseudonyms, and selective-disclosure proofs to enforce privacy-preserving access authorization, while leveraging Solidity-based smart contracts to realize automated and non-repudiable escrow settlement. By recording only pseudonymous usage evidence and hash-chained metering data on-chain, the system achieves strong unlinkability while preserving verifiable accountability and auditability. A full prototype demonstrates low end-to-end latency ($\approx$150 ms) and modest on-chain cost ($\approx$603K gas or $\approx$\$0.9), showing that SpexPay is practical for real-world DSS deployments. We also evaluated the user-side cryptographic operations on a Raspberry Pi 5 to assess scalability and suitability for edge-class hardware. Our code and artifacts are publicly available at https://github.com/iambarat/SpexPay.

Shiding Zhu, Yudi Qi, Yajie Wang et al.

Experience-driven self-evolution is critical for large language model (LLM) agents to improve through open-world interaction. However, existing experience learning methods mostly rely on single-agent loops, where the same agent executes tasks, summarizes outcomes, and determines memory content. This setup makes agents vulnerable to the Self-Confirmation Trap: wrong-but-self-consistent trajectories are misidentified as successful experience, leading to cumulative errors during retrieval and reuse. To address this issue, we propose EDV, an Execute-Distill-Verify framework for reliable experience learning. In the Execute stage, multiple heterogeneous agents explore the same task space in parallel to generate diverse candidate trajectories. In the Distill stage, a dedicated third-party agent comparatively analyzes these trajectories to produce candidate experiences, reducing executor-centric summarization bias. In the Verify stage, the execution group validates candidates via a consensus mechanism, and only approved experiences are written into shared or private memory. By decoupling the three stages, EDV transforms experience learning from isolated self-reflection into collaborative construction, filtering erroneous and noisy content before memory insertion. We evaluate EDV on three challenging long-horizon benchmarks: tau2-bench, Mind2Web and MMTB. Results show EDV consistently outperforms strong baselines, validating that reliable experience construction is essential for robust agent self-evolution. Our code is available at https://github.com/shidingz/EDV.

Shaoqiu Zhang, Maoquan Wang, Yuling Shi et al.

Large Language Model (LLM) coding agents have achieved strong results on software engineering tasks, yet repository exploration remains a major bottleneck: locating relevant code consumes substantial token budget and pollutes the agent's context with irrelevant snippets. In most agents, the same model explores the repository and solves the task, leaving exploratory reads and searches in the solver's history. We present FastContext, a dedicated exploration subagent that separates repository exploration from solving. Invoked on demand, FastContext issues parallel tool calls and returns concise file paths and line ranges as focused context. FastContext is powered by specialized exploration models spanning 4B--30B parameters. We bootstrap them from strong reference-model trajectories and refine them with task-grounded rewards for broad first-turn search, multi-turn evidence gathering, and precise citation generation. Across SWE-bench Multilingual, SWE-bench Pro, and SWE-QA, integrating FastContext into Mini-SWE-Agent improves end-to-end resolution rates up to 5.5\% while reducing coding-agent token consumption up to 60\%, with marginal overhead. These results show that repository exploration can be separated from solving and handled effectively by specialized models. Code and data: https://github.com/microsoft/fastcontext

Xiaolin Lin, Jingcun Wang, Olga Kondrateva et al.

Long-context large language model (LLM) inference is increasingly constrained by the memory footprint and decoding cost of key-value (KV) caches, limiting sustainable deployment on resource-constrained hardware. Existing KV cache eviction methods typically apply heuristic token scoring over all heads in GQA-based LLMs. These methods ignore the different functionalities of attention heads, leading to the eviction of critical tokens and thus degrading the performance of LLMs. To address this issue, we propose CompressKV, a resource-efficient KV-cache compression framework for GQA-based LLMs. Instead of aggregating attention scores from all heads, CompressKV identifies Semantic Retrieval Heads (SRHs) that capture both the initial and final tokens of a prompt and semantically important mid-context evidence, and uses them to select tokens whose KV pairs should be retained. Furthermore, CompressKV allocates cache budgets across layers according to offline estimates of layer-wise eviction error. Experiments on LongBench and Needle-in-a-Haystack show that CompressKV consistently outperforms existing KV-cache eviction methods across memory budgets. Notably, it preserves over 97\% of full-cache performance using only 3\% of the KV cache on LongBench question-answering tasks and achieves 90\% accuracy with just 0.7\% KV storage on Needle-in-a-Haystack. These results demonstrate an improved resource--performance trade-off for long-context LLM inference. Our code is publicly available at: https://github.com/TUDa-HWAI/CompressKV

Lichao Mou, Shilan Zhang, Chunlei Li et al.

Large vision-language models (LVLMs) can be adapted to specialized medical imaging tasks via parameter-efficient fine-tuning approaches such as low-rank adaptation (LoRA), leading to a growing ecosystem of expert models tailored to specific imaging modalities and clinical scenarios. However, deploying multiple expert LVLMs in practice incurs substantial computational and operational overhead. Model merging provides a promising solution by consolidating multiple experts into a single model without retraining, yet it remains largely unexplored in the medical domain. In this work, we present the first systematic study of model merging for medical LVLMs. We introduce MergeMedBench, a comprehensive benchmark spanning eight imaging modalities and diverse clinical task types, comprising 16 LoRA fine-tuned models built upon two mainstream architectures. We conduct an extensive evaluation of existing merging methods and further propose winner-take-all, a simple and hyperparameter-free approach that retains only the most dominant parameters across expert models. By preserving the critical parameters that govern model behavior and discarding weaker ones, our method avoids the information dilution inherent in averaging- or alignment-based strategies. Despite its simplicity, winner-take-all consistently outperforms existing approaches, offering both a new perspective on LoRA merging and a strong practical baseline for future research.

Jiaxi Liang, Yuanxiang Shi, Zezhou Yang et al.

Modern large-scale software systems often suffer from pervasive memory inefficiencies (e.g., bloat, churn), leading to excessive resource costs and performance degradation. Existing optimization workflows lack end-to-end automation, forcing developers to manually synthesize complex tool outputs into actionable and semantics-preserving fixes, precluding scalability in large codebases. To address this, this paper presents MOA, an LLM-driven framework that automatically detects and repairs recurring memory inefficiencies across production-scale codebases. Specifically, MOA operates through three agents: an Analyzer that mines anti-patterns from profiling data, a Checker Generator that synthesizes static analyzers through template-guided refinement, and a Patcher that generates optimization patches via state-machine-driven workflows. Our evaluation on OpenHarmony, an open-source operating system with over 100 million lines of C/C++ code, shows that MOA identifies 13 anti-patterns (9 previously unknown) from 3 profiled services, detects over 10,000 inefficiencies across a broader set of 7 services, and generates 769 patches with 92.5% expert acceptance rate, achieving 42.2% heap reduction and 10.6% binary size reduction on average. We envision MOA as a valuable tool for performance engineering at production scale.

Yao Shi, Kingfung Luo, Nan Tang et al.

Niche asset markets, such as Counter-Strike 2 (CS2) weapon skins, are small, volatile, and heavily driven by community discussions and platform rules. These properties make them hard for traditional quantitative models, but provide an ideal testbed for studying how large language models (LLMs) turn unstructured text into trading actions. We present CSTrader, a multi-agent framework for language-grounded trading in the CS2 skin market. The system first integrates heterogeneous signals from various sources, then uses specialized agents for technical analysis, liquidity, events, and (reversed) sentiment, and finally applies risk control, transaction friction, and portfolio management agents to produce buy, sell, or hold decisions under realistic trading frictions. We build a live-like evaluation environment with real CS2 data from a highly volatile period and evaluate several recent LLM backbones. Across models, CSTrader consistently outperforms both a falling market index (-15.62%) and simple single-prompt LLM baselines, achieving up to a 7.58% cumulative return with controlled risk. Ablation studies show that liquidity, reversed sentiment, and transaction friction agents are crucial for turning noisy language signals into stable profits, suggesting that niche, language-driven markets are a useful benchmark for future language-to-action research. Code is available at: https://github.com/IatomicreactorI/CSGOTrading?tab=readme-ov-file#quick-start

Tencent WorkBuddy Bench Team, Siqi Cai, Shaopeng Chen et al. · tencent-ai

We introduce Tencent WorkBuddy Bench, a multi-domain evaluation suite for coding agents; this report documents its construction methodology, scoring protocol, and a cross-model leaderboard. At its core is a unified evaluation framework for constructing and running distribution-informed coding-agent tasks across four work domains - Code, Web, Office, and Security. Rather than adapting public issue text, every task is reverse-engineered from a real commit, pull request, or business scenario and rewritten as a short, colloquial, role-played request, so that a task's prompt is not recoverable by web-searching the underlying issue, pull request, or commit thread. Because the dataset is released openly - task directories, environment images, evaluation harness, tests, and reference solutions - contamination resistance rests on this construction together with dataset versioning rather than on secrecy. The four subsets - repository-level engineering, front-end development, office and business workflows, and red-/blue-team security - probe complementary facets of real work, each with its own verification style. All are packaged in a uniform task-directory format and run, under a uniform and reproducible protocol, on two agent harnesses (CodeBuddy Code and Claude Code); the full open release makes the benchmark reproducible end to end and directly auditable, since any third party can re-run each task and inspect its content. Because each subset uses a different scoring instrument, scores are not comparable across subsets and the suite reports no suite-wide average. We report a cross-model leaderboard across several model families.

Ching-Hao Chiu, Hao-Wei Chung, Gelei Xu et al.

With the rapid adoption of generative AI, synthetic medical images pose growing risks, including diagnostic deception and insurance fraud. Although prior work has explored vision-language model (VLM)-based synthetic image detection, these evaluations typically consider images in isolation. In clinical practice, however, images are interpreted alongside structured records and metadata, and VLMs are increasingly deployed under joint image-record inputs. We uncover a previously underexamined multimodal vulnerability: when given both modalities, VLMs may overweight record context in authenticity judgments, such that the same image receives different predictions solely due to changes in its accompanying text. This raises concerns about robustness in real-world deployment. To systematically characterize this effect, we reformulate synthetic medical image detection as an audit of multimodal robustness at the image-record interface and introduce a paired benchmark that holds the image fixed while swapping controlled metadata variants. Across multiple imaging modalities, we evaluate diverse open-weight and frontier API VLMs and quantify how metadata alone shifts authenticity predictions. Our benchmark provides a standardized tool for assessing and improving multimodal robustness beyond image-only settings. The code is available at https://github.com/chiuhaohao/Beyond-Visual-Forensics.

Zongxia Li, Zhongzhi Li, Yucheng Shi et al.

AI agents have become capable of autonomously completing short, well-specified tasks. However, existing terminal benchmarks largely focus on simple problems that finish within minutes and are evaluated only by their final outcome. This setup overlooks intermediate progress and partial solutions, yielding sparse reward signals and an incomplete picture of agent capability. We introduce Long-Horizon-Terminal-Bench, a terminal benchmark of 46 long-horizon tasks spanning nine categories, including experiment reproduction, software engineering, multimodal analysis, interactive games, and scientific computing. Each task follows a Terminal-Bench-style setup with a reference solution or simulation engine, but is further decomposed into fine-grained graded subtasks. This design enables dense intermediate rewards and partial credit, allowing evaluation to capture not only whether an agent reaches the final goal, but also how far it progresses on open-ended workflows. Tasks in Long-Horizon-Terminal-Bench typically require hundreds of episodes and minutes to hours of execution, stressing long-horizon planning, long-context management, and iterative debugging rather than one-shot problem solving. We evaluate 15 frontier models and find that agents consume on average 9.9M tokens per task, with roughly 231 episodes and 85.3 minutes of execution time per run, making Long-Horizon-Terminal-Bench more demanding than prior terminal-based benchmarks. Even the strongest tested model achieves 15.2% pass@1 at a partial-reward threshold of 0.95 and 10.9% at a perfect-reward threshold of 1.0, while the mean pass rate across models is 4.3% and 1.7% under the two thresholds, respectively. These results reveal headroom for improvement. We further analyze failure modes and error patterns, and release Long-Horizon-Terminal-Bench to support future progress on long-horizon terminal agents.

Xinlong Chen, Jiafu Tang, Yue Ding et al.

Accurate and comprehensive video captions with consistent subject references are critical for downstream understanding and generation tasks. However, few existing benchmarks can objectively and comprehensively evaluate these properties across diverse durations and scenarios, thereby hindering the advancement of video captioning models. To bridge this gap, we propose CapRiCorn-1K, a comprehensive benchmark designed to evaluate both video captioning quality and subject referential consistency across long temporal horizons and diverse video domains. To accommodate varied evaluation needs, our benchmark supports both audiovisual and visual-only settings. Extensive experiments on CapRiCorn-1K reveal that current models generally struggle to generate accurate and comprehensive captions while maintaining consistent subject references. Moreover, as video duration increases, both the overall caption quality and subject referential consistency decline. Notably, our evaluation metrics exhibit strong correlations with the performance of downstream understanding and generation tasks conditioned on the generated captions, further validating their effectiveness. The project is available at https://github.com/xlchen0205/CapRiCorn-1K .

Suting Ni, Hanbing Zhang, Zhenyu Wei et al.

Tactile feedback is fundamental to Hand-Object Interaction (HOI), governing contact formation, force regulation, and stable manipulation, making it essential for achieving true human-like dexterous manipulation. Yet, current human-to-robot dexterous transfer pipelines primarily rely on kinematic trajectories, resulting in motion imitation without physically grounded interaction. To address this, we introduce TactiDex, a real-world tactile-guided benchmark specifically designed to move dexterous manipulation beyond kinematic mimicry toward contact-level human-likeness. TactiDex provides a comprehensive dataset that elegantly aligns whole-hand tactile signals with multi-granularity kinematic and object states, coupled with standardized evaluation metrics. Building upon this data paradigm, we propose a tactile-driven transfer framework that effectively translates human demonstrations into physically plausible robotic execution. We introduce TactiSkill, a framework built upon a novel tri-component tactile reward that innovatively uses tactile signals as structured supervision. This reward unifies guidance, human-like alignment, and contact constraints into a single objective. Through comprehensive experiments on both single and bimanual tasks, we demonstrate that TactiSkill achieves superior performance in manipulation success and physical realism. This work lays a crucial foundation for advancing tactile-aware dexterous manipulation. Our project page at https://tactidex.github.io/.

Yunao Huang, Shiyu Sang, Haotao Lu et al.

Contact-rich robot manipulation requires physical interaction cues that are often invisible to cameras, making tactile sensing essential for robust control. However, scaling visuo-tactile robot learning remains difficult because real tactile interaction data are expensive to collect, hardware-dependent, and limited in task and scene diversity. We present ViTacWorld, an action-conditioned visuo-tactile world model for scalable contact-rich robot manipulation. ViTacWorld leverages public real tactile datasets and a constructed simulation environment to scale visuo-tactile-action data, exploiting the fact that tactile signals are directly grounded in physical contact and can exhibit a smaller simulation-to-real gap than purely visual observations. The model is first pretrained with large-scale real and simulated visuo-tactile trajectories, and then finetuned with real-world policy rollouts to better match downstream manipulation behaviors. Given robot actions, ViTacWorld predicts temporally aligned visual observations and tactile feedback, enabling visuo-tactile-action rollout generation. To the best of our knowledge, ViTacWorld is the first framework that uses a world model for robot visuo-tactile-action trajectory generation and policy evaluation. It serves two roles: synthesizing rollouts to improve downstream tactile policies, and evaluating policies by predicting action-conditioned visuo-tactile outcomes under controlled action sequences. Experiments on contact-rich manipulation tasks show that ViTacWorld generates physically meaningful rollouts, improves policy performance through scalable data augmentation, and enables action-conditioned policy evaluation. Project page: https://vitacworld.github.io/

Xinlei Yu, Gen Li, Qingyi Si et al.

On-policy distillation (OPD) offers superior capacity transfer by supervising student-sampled trajectories with dense token-level signals. To furnish high-quality supervision sources and thereby elevate the performance frontier of distillation, an intuitive direction is to infuse privileged information to either teacher or student itself. However, this additional input induces a potential failure mode we dub privilege illusion: a pattern that conflates the transferable capability gap that students are meant to close, and the information asymmetry gap that can only be mimicked but never replicated. This issue is further amplified by the inherent non-uniformity of token-level supervision, where only a small subset of tokens carries pivotal capability-bearing signals. To this end, we propose DOPD, an advantage-aware dual distillation paradigm that dynamically routes token-level supervision between privileged teacher and privileged student policies based on their advantage gap and relative probabilities. Each token receives supervision of different strength, objective, and strategy from either teacher or student itself, which transfers credible capability while simultaneously receiving auxiliary signals, to alleviate privilege illusion. Extensive experiments on both large language model (LLM) and vision-language model (VLM) settings demonstrate that DOPD consistently outperforms Vanilla OPD and other counterparts. Further results on stability, robustness, continual learning, and out-of-distribution tasks validate its superiority.

Shaoteng Zhang, Weiwei Cao, Wanxing Chang et al.

Medical images require comprehensive and accurate interpretation to support the diagnosis of diverse clincial conditions. Recent vision-language generalist models offer broad task coverage and promising zero-shot capabilities, yet often lack fine-grained anatomical and lesion awareness for reliable diagnosis and spatial interpretability. In contrast, supervised specialist models achieve strong performance on specific tasks but typically lack generalization across diseases and anatomies. In this work, we present SuG, a Super-Generalist framework that unifies generalist vision-language learning with specialist objectives, enabling both broad generalization and specialist-level diagnostic capability. We perform specialist-enhanced vision-language alignment in SuG by incorporating spatial priors from multiple segmentation experts, including anatomy, class-specific lesion and class-agnostic lesion segmentors that captures lesions beyond anatomies annotated during training. To improve lesion grounding capability, we leverage lesion masks as spatial priors to calibrate text-conditioned visual attention, encouraging disease-related semantics to focus on clinically relevant regions. We evaluate SuG on extensive chest and abdominal CT benchmarks, including CT-RATE, Merlin, MedVL-CT69K, and several in-house tumor datasets. SuG achieves state-of-the-art performance across a wide range of disease diagnosis tasks and surpasses specialist models on several critical tumor diagnosis benchmarks. Furthermore, SuG demonstrates strong lesion grounding capability, including robust generalization to lesion types lacking class-specific supervision.

Junyu Dai, Xinyue Fan, Weiqin Li et al.

In this report, we present a unified song generation framework capable of producing high-quality full-length music from lyrics, text descriptions, and musical attributes. The proposed framework supports three tasks: Lyrics-to-Song Generation, which generates complete songs from text descriptions, lyrics, and musical attributes; Instrumental Music Generation, which creates music without vocals; and Cover Song Generation, which reinterprets existing songs with different styles while preserving their melodic content. Architecturally, our system consists of four main components: a semantic-aware tokenizer, hybird-LM, FullDiT, and a two-level melody module. The tokenizer encodes audio into 8-codebook RVQ tokens for efficient discrete music representation. Based on these tokens, hybird-LM performs hierarchical autoregressive audio-token modeling for full-song generation. To improve audio fidelity, FullDiT performs full-song flow matching in a continuous VAE latent space conditioned on codec tokens, lyrics, and text captions. For cover song generation, the melody module extracts and discretizes melody cues from reference audio to guide generation while preserving the original melodic content. Finally, we investigate DPO, GRPO, and OPD as reward-based post-training strategies for hybird-LM and apply flow-based GRPO to FullDiT to improve musicality and rendering quality. Experimental results on a multilingual automatic benchmark, complemented by the Artificial Analysis Music with Vocals leaderboard, show that the proposed framework achieves competitive performance in the evaluated settings.

Yuhang Wang, Yuling Shi, Shaoqiu Zhang et al.

Pruning long context for coding agents has been a vital technology for efficient context management. While existing context pruning methods such as SWE-Pruner realize this by attaching a separate code classifier, we find the agent itself encodes internal representations indicating the relevance of code context when reading tool output. Based on this finding, we propose SWE-Pruner Pro, which prunes tool outputs directly inside the agent. Concretely, a small head turns the agent's own internal representations into a keep-or-prune label for each line, with a length-aware embedding keyed to each tool output's line count. Across two open-weight backbones and four multi-turn benchmarks, SWE-Pruner Pro saves up to 39% of prompt and completion tokens while preserving task quality, with bounded inference overhead. Notably, on MiMo-V2-Flash SWE-Pruner Pro additionally raises the SWE-Bench Verified resolve rate by +3.8% and the long-context Oolong accuracy by +2.2 points.

Lianghua Huang, Zhifan Wu, Wei Wang et al.

We present Wan-Streamer, a native-streaming, end-to-end interactive foundation model designed from the ground up for real-time, low-latency, full-duplex audio-visual interaction. Wan-Streamer seamlessly models language, audio, and video as both input and output within a single Transformer, where the sequence is represented as interleaved visual, audio, and text input tokens together with visual, audio, and text output tokens, coordinated by block-causal attention for incremental streaming. Unlike cascaded interactive systems that rely on separate VAD, ASR, language, TTS, audio-driven animation, or video-generation modules, Wan-Streamer does not rely on external language, speech, avatar, or video-generation modules: perception, reasoning, generation, response timing, turn management, and cross-modal synchronization are learned jointly within one unified model, reducing pipeline latency and error accumulation. To support natural audio-visual responsiveness, we redesign the entire stack around streamability, including causal encoders, causal decoders, block-causal attention, and low-latency multimodal token scheduling, enabling streaming units as short as 160 ms at 25 fps. Wan-Streamer achieves approximately 200 ms model-side response latency and approximately 550 ms total interaction latency when combined with 350 ms bidirectional network latency, supporting sub-second duplex audio-visual communication. These results position Wan-Streamer as a unified, end-to-end, multimodal interactive foundation model for low-latency streaming interaction.

Yipeng Shi, Zhipeng Ma, Yue Wang et al.

In long-horizon LLM agent reinforcement learning, weak policies often repeat similar failures, producing uninformative rollout trajectories and limiting effective policy optimization. Existing skill-centric methods improve exploration by optimizing, filtering, or internalizing reusable skills. However, they remain centered on the skills themselves rather than being designed as adaptive training-time support for the evolving policy. To address this, we propose a policy-centric training paradigm that reframes skills as a dynamic training scaffold. Our framework, Pats, converts rollout groups from the latest policy into evidence cards and uses task-specific evaluation to adjust the context used in subsequent rollouts. Concrete guidance helps weak policies to complete challenging tasks. As policy improves, redundant context is revised or removed to reduce reliance on explicit guidance while preserving useful rollout variation. The policy is optimized with environmental rewards using standard RLVR, and the training scaffold is discarded at deployment. On ALFWorld and WebShop, Pats improves over strong baselines by up to 18.6%. Across seven search-augmented QA benchmarks, it remains competitive while using 32.1% fewer prompt tokens than the baseline.

Hongnan Ma, Yiwei Shi, Mengyue Yang et al.

Faithful explanations of time-series classifiers should identify subsequences that are not only sufficient to preserve a black-box model's prediction, but also necessary for maintaining it. However, existing sufficiency-oriented methods can assign high importance to spurious subsequences that support the prediction without being essential to the model's decision. We introduce \textbf{TimePNS}, a necessity-aware framework for time-series explanation. Inspired by Pearl's counterfactual notion of necessity, TimePNS assesses whether a temporal factor is necessary by intervening on it and measuring whether the original prediction is disrupted. The framework adopts a two-stage design. Stage I learns an identifiable causal generative process together with a sufficiency-oriented explanation mask. Stage II performs counterfactual interventions on temporal factors to derive necessity signals, which supervise a temporal gate that refines the initial explanation by suppressing non-essential components and emphasizing counterfactually necessary ones. Experiments on synthetic and real-world time-series benchmarks show that TimePNS more accurately identifies decision-critical subsequences and consistently improves sufficiency-necessity trade-offs over strong baselines.

Lianghua Huang, Zhi-Fan Wu, Yupeng Shi et al.

We present Wan-Streamer v0.3, which reframes our native-streaming interaction model under a single organizing view: a video is a world plus an event stream. The world is the persistent context in which a video unfolds, including the environment, scene, subjects, ambient acoustic conditions, voice characteristics, and other relatively stable conditions. The event stream is everything that changes over time within that world, including scene or environmental changes, subject behavior, speech, and other sounds. This yields a general-purpose pretraining task over large amounts of real video: given a world and incoming input, predict how the world moves, changes, and responds in real time. The resulting competence can be specialized to a broad family of real-time downstream tasks. We instantiate it on real-time full-duplex audio-visual interaction, where the event stream is the agent's speech together with free-form behavior. Functionally, the model's multimodal understanding process is vision-language-action-like: it maps multimodal user input to language-form speech and behavior actions. Wan-Streamer v0.3 preserves the v0.2 operating point: 640x368 video at 25 FPS, a 160 ms streaming unit, approximately 200 ms model-side response latency, and approximately 550 ms total interaction latency under a 350 ms bidirectional network budget.

Yichen Shi, Yuzhi Liu, Zhuofu Tao et al.

A rich and recognizable component library is the cornerstone of printed circuit board (PCB) design and generation. Traditionally, engineers manually create symbols and footprints and design PCB schematics, which is time-consuming and error-prone. Leveraging multimodal large language models (MLLMs), we develop SFgen, an agentic recognition and generation flow of symbol and footprint for electronic components. SFgen achieves 86% accuracy for symbol generation and 80% accuracy for footprint generation. We use the SFgen method to create SFnet, a database of symbols and footprints. It now has 1000 components and is expanding constantly, which lays the foundation for automatic generation of PCB designs.

Hang Ye, Xinyan Jiang, Yuedong Shi et al.

Modern power systems increasingly require probabilistic forecasts amid interacting uncertainties from renewable intermittency, flexible demand, market volatility, and weather-dependent generation. However, existing methods often treat multi-scale decomposition, exogenous-variable alignment, and probabilistic output as separate steps, obscuring how predictable structures and uncertainty-bearing fluctuations jointly shape the forecast distribution. This paper proposes a state-space exogenous-context and temporal-frequency resolution architecture for general probabilistic energy forecasting. Its central premise is that trend-periodic components primarily determine the baseline trajectory, whereas high-frequency residuals and external perturbations govern the spread and asymmetry of forecast uncertainty. Accordingly, the architecture adaptively separates deterministic and residual streams, aligns exogenous context with both, refines the deterministic backbone through multi-resolution spectral-temporal state-space modeling, and estimates ordered quantile boundaries from their complementary representations. Experiments on load, price, solar, and wind forecasting achieve the best continuous ranked probability score in 14 of 18 settings, reducing average CRPS by 5.74\% and upper-tail quantile risk by 7.27\% over the strongest baselines. These results support deterministic-stochastic separation as an effective design principle for general probabilistic energy forecasting.

Dongjian Ma, Silin Chen, Yufei Yang et al.

Coding agents powered by large language models have demonstrated strong performance on software engineering tasks. Yet most agents consume repositories almost entirely as text, which differs from how human developers use visual structure such as folder hierarchies and dependency relationships to orient themselves in large codebases. With multimodal large language models (MLLMs), it is an open question whether agents can effectively benefit from visual representations of repositories. This paper presents the first systematic empirical study of visual repository representations for LLM-based agents on repository-level issue resolution. We evaluate four recent multimodal models. Our results show that a strictly vision-only setup degrades accuracy and increases token cost, because agents lack sufficient symbolic detail and compensate with repeated visual queries. In contrast, integrating visual graphs of repository structure as a supplementary modality alongside standard text interfaces helps agents understand structure more efficiently: input token consumption decreases by up to 26% while issue-resolution accuracy is maintained or improved. Visualization is most useful during fault localization and when the agent autonomously controls exploration depth. These findings point to a practical hybrid text-and-vision design for next-generation coding agents.

Xirui Kang, Yanpei Shi, Lucy Liang et al.

Modern Vision-Language-Action (VLA) models must bridge pretrained vision-language reasoning and precise continuous robot control. Existing action tokenizers discretize actions primarily for reconstruction, producing codes that preserve motion geometry but provide only weak semantic supervision to the backbone. We therefore formulate action tokenization not as mere compression, but as semantic interface learning between multimodal reasoning and executable control. To this end, we introduce X-Tokenizer, a lightweight encoder-Semantic Residual Quantization (SRQ)-decoder architecture that provides a shared action interface across diverse robotic arm embodiments. Its key component, SRQ, imposes an asymmetric structure on residual vector quantization: the first level is trained with Masked Action Modeling (MAM) to form a discrete action language that captures coarse motion intent, while deeper levels remain reconstruction-oriented residuals that preserve fine-grained details. To further align action tokens with multimodal semantics, X-Tokenizer is pretrained with contrastive alignment to the representation space of a pretrained foundation model and with next-frame vision-language feature prediction. Pretrained on 2.4M trajectories (2.0B action frames), a single frozen X-Tokenizer plugs into a mixed discrete-continuous VLA as a representation-shaping supervision signal. X-Tokenizer achieves top real-world aggregate and strong RoboTwin 2.0 simulation results. Outperforming FAST in multimodal grounding (+13.5%) and long-horizon tasks (+8.25), it shows that action tokenizers serve as semantic interfaces for VLA pretraining beyond mere action compression.

Peng Ren, Haoyang Ge, Jiang Zhao et al.

Vision-language-action policies are a promising foundation for general robot control, but long-horizon humanoid loco-manipulation requires the robot to treat task objects as persistent physical entities across movement, contact, occlusion, and recovery. We study this problem as object-state divergence: the object state used to condition a whole-body action can differ from the state used to decide whether the action achieved the intended physical relation. We propose \emph{Persistent Object Tokenization} (POT), which maintains role-indexed 3D object records from RGB-D observations and converts them into object tokens for a whole-body action expert. Instantiated as \emph{POT-VLA}, the same object records condition action generation and support geometric predicate checks, yielding a closed-loop execution system in which object state is both actionable and verifiable. On a Unitree G1, POT-VLA improves a matched direct GR00T-N1.7 baseline from 39/80 to 71/80 successes over eight real-world task families. In an external Being-0-aligned reference, POT-VLA achieves 44/50 successes on aligned service tasks, compared with the 37/50 success reported by the Being-0 paper. The largest gains occur on tasks requiring maintained 3D relations, suggesting that persistent object-centered state is a useful abstraction for verifiable humanoid VLA execution.

Siqi Yan, Jiebao Zhang, Xi Yao et al.

The surge of GPU-intensive workloads in artificial intelligence (AI) data centers drives massive energy demands, leading to soaring costs and significant stress on local power distribution networks. Coordinating delay-tolerant workload scheduling with power grid conditions via precise workload prediction can mitigate these issues. However, a critical gap remains in conventional approaches, i.e., minimizing prediction error does not necessarily lead to minimized downstream operational loss. Hence, this paper proposes an end-to-end Predict-Then-Schedule (PTS) framework that integrates upstream workload prediction with downstream scheduling optimization. By leveraging differentiable convex optimization, the PTS framework maps input features directly to optimal scheduling and enables gradient-based training. Furthermore, to respect the data center's capacity, a workload over-shifted loss combining electricity cost with a penalty for load-shedding is introduced to evaluate scheduling quality. Experiments demonstrate that the proposed framework significantly reduces operational cost and enhances system security compared to the conventional two-stage baseline.

Daniela Rojas, Abdulwahab Albassam, Aidan G. Leung et al.

Large language models (LLMs) and agentic AI systems have evolved from natural language tasks to using external tools to plan, retrieve, and act in technical domains. In smart grids, recent work applies agentic schemes to forecasting, optimization, and control, wrapping trusted solvers behind language interfaces and orchestrating multi-step workflows. The literature lacks a unified approach to designing and evaluating such systems. LLMs can produce numerically plausible yet physically infeasible outputs, evaluation protocols vary across tasks, and the boundary between what the model should and should not compute is implicit. This paper presents a solver-grounded design principle: a numerical result is reported only when it originates from a trusted tool and passes explicit verification. We review the building blocks of LLM and agentic AI systems for power systems: prompting strategies and agentic architectures. We instantiate the principle in four case studies: wind power forecasting, EV charging scheduling, power flow analysis, and contingency diagnosis, each comparing an LLM-only baseline against its solver-grounded counterpart on identical data and metrics. EVAgent reproduces the CVXPY optimum while reducing LLM-only unmet energy by 7.5-9.5x, and GridDebugAgent repairs 17/39 contingency cases while reducing total violations by 52.3%. We propose a four-group evaluation framework spanning task utility, solver-grounded correctness, faithfulness and safe failure, and cost and latency. A consistent division of labor emerges: the agentic system reliably orchestrates, retrieves, and explains, while trusted tools compute and a verification gate decides what is reported.

Zhenjia Li, Jinrang Jia, Yifeng Shi

A single panorama captures the full visual sphere from one camera center, yet confines users to looking around in place without enabling true scene exploration. Converting a single panorama into a persistent, renderable 3D representation for free-viewpoint navigation has attracted growing interest; existing methods either adopt iterative per-view completion that propagates inpainting results to update the underlying geometry, leading to progressive error accumulation and cumbersome multi-step pipelines, or leverage the temporal consistency priors of video generation models, yet the continuous-trajectory constraint intrinsic to such models limits their flexibility in covering scenes from multiple directions simultaneously. We present Pano2World, which takes a single indoor panorama as input and directly outputs a persistent, explorable 3D Gaussian scene. Given the source panorama, Pano2World first reconstructs a coarse 3D Gaussian proxy and renders it at adaptively sampled nearby poses to obtain geometrically aligned guidance panoramas; a panoramic diffusion model then jointly denoises all target views via View-Aware Attention Routing, where each target view simultaneously receives geometric constraints from its corresponding guidance panorama and global semantic guidance from the source panorama, naturally enforcing cross-view consistency. To avoid the information loss incurred by decoding the multi-view hidden features formed during joint denoising back to the pixel domain via VAE, we introduce Latent Feature Adapter, a geometry-aware bridge module that directly distills these hidden features into a scene latent, subsequently decoded into the final 3D Gaussian scene. Experiments demonstrate that Pano2World significantly outperforms existing methods on the multi-position panoramic novel-view synthesis benchmark.

Tingwei Chen, Yunxiao Shi, Zhengdong Chu et al.

Open agent marketplaces list native agents, tool bundles, and reusable skill packages in the same search interface, yet practitioners still have little guidance on how to retrieve across this mixed catalog. We study whether off-the-shelf retrieval models, trained for general text retrieval, can be adapted to match user queries to executable agent capabilities, and whether the learned signal transfers beyond the benchmark used for tuning. We fine-tune three open retrieval backbones, BGE-base, KaLM-v1.5, and EasyRec, on AgentSelect, which represents marketplace-visible units as capability profiles derived from public metadata, and test transfer on two catalogs not seen during training: MuleRun native agents and a ClawHub benchmark of 50 skills with 1,000 queries. Adaptation helps on both catalogs. Code and data will be released upon publication.

Renqin Cai, Dawei Sun, Yuanjun Yao et al.

As scalability becomes increasingly important in recommendation modeling, recent architectures have advanced the modeling of two broad sources of ranking signals along separate paths: non-sequence features, including user, item, context, and cross features; and sequence features from user behavior histories. Wukong and HSTU have emerged as representative scalable backbones for these paths: Wukong scales high-order non-sequence feature-interaction modeling, while HSTU scales long user-behavior sequence modeling. Despite their complementary strengths, practical architectures that combine these two types of feature modeling remain underexplored. We present WHALE, a scalable unified recommendation architecture that jointly models non-sequence and sequence features on top of Wukong and HSTU. Each WHALE layer contains a Wukong module, an HSTU module, and an attention-based fusion module in which Wukong-derived interaction representations query HSTU-derived behavior representations. This design keeps both backbones active throughout the network and enables progressive Wukong-HSTU exchange, allowing high-order feature crosses to repeatedly retrieve fine-grained evidence from long user histories. To make WHALE practical for industrial deployment, we introduce customized Triton kernels and other model-systems co-design techniques to improve training and inference efficiency. On large-scale industrial recommendation data, WHALE achieves consistent gains in offline experiments. Additionally, it delivers positive online gains with a modest serving-throughput trade-off. The method has been deployed in production systems. Overall, WHALE provides a practical example of how these two sources of information can be scalably unified in an industrial recommendation model.

Yang Shi, Jiaheng Fu, Yihe Huang et al.

Large language models (LLMs) are increasingly deployed as autonomous agents that interact with external tools and services via the Model Context Protocol (MCP), a standardized interface for dynamic tool invocation. While MCP simplifies integration, it also expands the attack surface and enables generic exploits across multiple servers. Despite prior work on malicious MCP servers, the vulnerability landscape of MCP servers remains underexplored. In this work, we systematically analyze MCP server vulnerabilities, focusing on metadata characteristics, vulnerable code patterns, and community responses. Our study reveals that taint-style vulnerabilities constitute a substantial fraction of MCP server vulnerabilities, require significant code modifications to remediate, and are met with slow community responses. Motivated by these findings, we propose SPELLSMITH, presenting a novel textbased avenue for shielding taint-style vulnerabilities in MCP servers. In particular, SPELLSMITH analyzes the high-risk capabilities exposed by an MCP server and combines them with tool descriptions and parameter semantics to identify potential taint-style vulnerability risks, thereby constructing a tool-level risk profile. Then, SPELLSMITH leverages the Description property of the protocol to embed behavioral guidance (Description Enhancement Module) and exploits LLMs' self-reflection capabilities (Self-Reflection Module) to iteratively evaluate and refine outputs. By strengthening LLM internal decision-making, SPELLSMITH provides an active and unified mitigation strategy that generalizes across multiple vulnerabilities, reducing reliance on context-specific code-level fixes. Our experiments demonstrate that SPELLSMITH effectively mitigates taint-style vulnerability exploitation in MCP servers, highlighting its practical applicability and advantages over traditional code-level mitigations.

Lianghua Huang, Zhi-Fan Wu, Yupeng Shi et al.

We present Wan-Streamer v0.2, a latency-preserving upgrade of the native-streaming, end-to-end audio-visual interaction model. v0.2 keeps the v0.1 modeling formulation, but raises the interactive output stream from 192x336 to 640x368 while preserving approximately 200 ms model-side signal-to-signal latency at 25 FPS. The higher-resolution stream supports scene-grounded mid-shot agents whose posture, gaze, hands, nearby objects, and local scene layout remain legible during real-time conversation. To support the larger visual stream without adding user-visible delay, v0.2 keeps the thinker as a single-GPU low-latency path for streaming perception, the short language/state Transformer pass that builds the generation cache, and final decoding. The performer becomes a multi-GPU Ulysses-style context-parallel group for the expensive next-unit latent generation. Each performer rank writes incoming K/V into a pre-sharded local cache. The long high-resolution latent video sequence is split across ranks for denoising and gathered through Ulysses communication, while the much shorter audio latent sequence is generated without sequence sharding. In this split, the thinker's language/state computation reaches the performer only as K/V conditioning, so no separate language sequence has to be communicated inside the performer group. This concentrates additional hardware on visual generation while preserving the compact thinker-performer boundary, keeping total remote interaction latency at approximately 550 ms when a 350 ms bidirectional network budget is included.

Yuan Xu, Youheng Shi, Chengyang Li et al.

Vision-Language-Action (VLA) models inherit rich semantic representations from pretrained Vision-Language Models, yet fine-tuning on limited robot demonstrations degrades this structure and undermines generalization. A fundamental question therefore arises: what constitutes a good action representation? Inspired by the mirror neuron theory's insight that observation and execution share an intention-level encoding, we examine whether a robot's action representations preserve the semantic structure captured by pretrained encoders. Systematic probing confirms that this structure erodes during finetuning, and that its quality synchronizes with both task success and out-of-distribution generalization. We further introduce a plug-and-play method that anchors action representations to a semantic manifold while decomposing representations into a shared semantic channel and a private channel, all discarded at inference, leaving the deployed model unchanged. Validated on different VLA backbones across simulation and real-world benchmarks, our method yields up to +18.7% on real-world in-distribution tasks and +21.5% on out-of-distribution generalization.

Likai Pei, Jiahao Zheng, Xueji Zhao et al.

Probabilistic computation plays an important role in trustworthy edge intelligence to quantify uncertainty, enhance robustness, reconstruct data, and protect privacy, but its adoption is limited by the orders-of-magnitude data throughput gap between Gaussian random number generation (GRNG) and computation, as well as instruction overhead. This paper introduces probabilistic memory (p-MEM), a unified memory primitive that stores distribution parameters, such as mean and standard deviation, and samples directly at the native memory bandwidth, where deterministic data becomes the zero-variance special case. Using a layout-validated p-MEM simulator, we comprehensively explore device choices, memory specifications, and technology nodes, showing that p-MEM can achieve more than 1000 GSa/s/mm^2 GRNG throughput, including memory-array access. Integrated into CPU/GPU systems, p-MEM reduces instruction count by up to 2.19x/4.37x, sampling latency by 562x/3.45x, and energy by 295.5x/3.53x for Bayesian neural network workloads, providing a scalable hardware substrate for trustworthy probabilistic AI.

Dayu Xia, Yue Shi, Yao Mu et al.

Vision-language models (VLMs) are increasingly explored as visual critics, reward generators, and failure detectors in robotic manipulation. These roles implicitly require models to judge not only final task success, but also how a manipulation execution is physically and temporally progressing. However, existing evaluations fail to test whether VLMs possess fine-grained process understanding. To address this gap, we present RoboProcessBench, a benchmark for process-aware understanding in vision-language robotic manipulation. RoboProcessBench decomposes such capability into two complementary dimensions, \emph{static monitoring} and \emph{dynamic reasoning}, instantiated as 12 diagnostic question families covering phase, contact, motion, coordination, primitive-local progress, temporal order, outcome, and primitive-level transitions. Built from physically grounded execution traces, the curated benchmark corpus ProcessData contains \textasciitilde 58k question-answer pairs across 260 manipulation tasks, which is further split into ProcessData-SFT and ProcessData-Eval for post-training and evaluation purposes. Extensive evaluation of various VLMs on ProcessData-Eval reveals broad limitations across 12 diagnostic task families, suggesting current models still lack robust process-aware understanding of manipulation executions. But with ProcessData-SFT, the post-trained \textit{Qwen2.5-VL-7B} and \textit{InternVL-3-8B} exhibit consistent gains on local state, motion, progress, and primitive-aware cues. These results demonstrate that RoboProcessBench serves as both an evaluation benchmark and a learnable supervision source for developing VLMs capable of monitoring and evaluating robotic manipulation processes. Project webpage: \href{https://processbench-2026.github.io/RoboProcessBench-Web/}{https://processbench-2026.github.io}.

Chang Liu, Jiaqi Liu, Zhoutong Ye et al.

We present AA, a multi-view multimodal dataset for screen-based gaze estimation. The dataset captures synchronized facial observations from eight fixed screen-mounted cameras and two additional side-view cameras, paired with precise screen-space gaze targets collected under controlled fixation conditions. Each sample contains multi-view face observations together with structured facial region crops, enabling multimodal learning from both global and local visual cues. Unlike existing single-view gaze datasets, AA provides multi-view coverage from both screen-mounted and side-mounted perspectives, enabling more robust modeling under viewpoint variation and occlusion. The dataset includes subject-independent evaluation splits and a standardized data processing pipeline to support reproducible research in gaze estimation.

Chao Wang, Hongtao Tian, Tao Yang et al.

Group Relative Policy Optimization (GRPO) is a default recipe for process-supervised reinforcement learning of LLM reasoners, and dense process supervision -- via learned process reward models (PRMs) or on-policy-distillation KL signals -- is a common way to densify its otherwise weak outcome reward. Layering such a step-level signal on top of GRPO's group-standardized advantage, however, exposes three structural pathologies: \emph{channel contamination} between the pooled process, outcome, and format streams at group standardization; \emph{resolution mismatch} between the granularity of the process signal and the granularity of the logical decisions being credited; and a \emph{cumulative trap} by which GRPO's return-to-go sum surfaces either length inflation or truncated exploration depending on the sign regime of the signal. We propose \textbf{PASS} (\emph{Process Advantage Signal Shaping}), a compact middleware that sits between any scalar step-level process signal and GRPO's clipped surrogate and addresses the three pathologies in turn: \emph{Advantage Fusion} standardizes the three streams independently within each group, \emph{Chunk-by-Value} derives value-homogeneous chunks from the signal itself and broadcasts credit within each chunk, and \emph{Divide-Length} converts the cumulative objective into an average-value-density score. We validate PASS across two domains and two process-signal paradigms -- a learned PRM on mathematical reasoning and an on-policy-distillation KL signal (with a generalized variant) on multi-hop question answering -- and under two group-standardization operators. In every regime PASS delivers a consistent pass@1 gain over the corresponding GRPO baseline.

Zhi Chen, Zhensu Sun, Yuling Shi et al.

Repository-level performance-optimization benchmarks such as GSO, SWE-Perf and SWE-fficiency evaluate coding agents by applying patches to real repositories and comparing runtime against unoptimized baselines and official reference patches. Their leaderboard scores are increasingly used as evidence of coding-agent progress, but those scores can conflate runtime instability, benchmark-specific scoring rules, and how many tasks are already solved by at least one public submission. We audit these issues across the three benchmarks. First, we replay the official reference patches for 740 code optimization tasks across four common types of Google Cloud machines. Most benchmark tasks can be replayed, but their reference patches satisfy the original benchmark validity rules in every cross-machine replay for only 39/102 GSO tasks, 11/140 SWE-Perf tasks, and 411/498 SWE-fficiency tasks; SWE-Perf is especially fragile because many reference patches produce close-to-zero runtime changes. Second, we show that public submission rankings depend strongly on the benchmark scoring rule. Among eight public submissions shared by GSO and SWE-fficiency, the official rankings disagree on 9 of 28 pairwise submission comparisons, and SWE-fficiency's leaderboard scoring rule assigns the worst ten tasks overly high score weights of 58.5%-82.8%. Third, looking across 10 public submissions for each task, we find that at least one submission matches or beats the reference patch on 85.3% (384/450) of replay-valid GSO and SWE-fficiency tasks, and beats the unoptimized base code on 99.8% (449/450). Our study complements leaderboard scores by identifying tasks with more reliable performance signals, quantifying per-task score contributions, and exposing the remaining performance gaps that are hidden by aggregate rankings.