Searcharxiv⌕ Search

arXiv subjects

Wenhao Yan

Publications and source records attributed to Wenhao Yan.

12 recordsLinked to original sources
arXiv2026

How Much Regularization Survives Averaging? Update Masking in Federated Learning

Federated learning on non-IID data seeks flat minima to generalize across clients, and existing methods borrow sharpness-aware minimization from centralized training. There is a second way to reach flat minima, in which the regularization comes for free from noise added to the parameter updates, and it has never been carried over to the federated setting as an implicit regularizer. We show the reason. Masking charges the optimizer for moving in sharp directions. We prove that when each client draws its own mask, federated averaging weakens that charge by exactly the cohort size, and that giving every client the same mask brings it back by a factor equal to the inverse gradient diversity of the cohort. In our experiment setting on CIFAR-10, that factor is 1.19 out of a possible 10. Turning off minibatch sampling raises it to 8.96, while changing data heterogeneity a thousandfold leaves it between 1.17 and 1.50. The configurations keeping the regularization train far too poorly to use.

cs.LG↗
arXiv2026

AutoDesign: Meta-Harness Optimization for Long-Horizon Agentic Design

Transforming multimodal sources into condensed and structured media outputs can be fundamentally conceptualized as a long-horizon agentic process centered on a model-harness system. While an ideal harness system should align with human design priors and accumulate reusable experience through empirical exploration to drive recursive self-improvement, existing paradigms remain static and fall short of this capability. In this paper, we present AutoDesign, a framework that aligns with human design priors, where a meta-harness optimizer guides a code agent to recursively improve harness based on rollout feedback. To instantiate and evaluate this framework, we focus on the academic paper-to-poster generation task and introduce PosterBench, comprising a 100-paper Main Track spanning five disciplines and PosterBench-mini, a shared 10-paper subset for controlled evaluation. On the PosterBench Main Track, AutoDesign achieves the highest score of 78.32, surpassing the closed-source commercial system Claude Design by 7.45 points. Across seven controlled code-agent-model configurations, integrating the learned DesignHarness consistently improves performance, increasing the average PosterBench Score from 54.99 to 67.39 (+12.4%). In a fully autonomous long-horizon loop, it executes 253 tool calls and 11 editing turns within 40 minutes for under $3, reaching average conference-poster quality in human evaluation. A system-blind human study further demonstrates that AutoDesign achieves the highest human preference among evaluated systems.

cs.CV↗
arXiv2026

VideoCoCo: Code-as-CoT for Physically-Consistent Video Generation via an Agentic Dual-Engine System

Text-to-video models have achieved remarkable visual quality, yet they still struggle to generate physically consistent dynamics because the temporal evolution of a scene must be inferred implicitly from a highly compressed text prompt. Existing chain-of-thought approaches introduce intermediate plans or visual states, but these representations are typically non-executable or temporally sparse, limiting their ability to instantiate and control the complete spatiotemporal process. To address this limitation, we introduce VideoCoCo, an agentic dual-engine framework in which executable Blender code serves as a process-level chain of thought. Given a text prompt, a coding agent synthesizes a Blender program that explicitly specifies the scene and its temporal evolution. The executable simulation engine runs the program to produce a deterministic spatiotemporal draft, which is subsequently transformed into a photorealistic video by a generative video engine through draft-conditioned editing. This decomposition separates process-level reasoning from high-fidelity visual realization. To adapt the video editor to simulated drafts, we construct VideoCoCo-3K, a curated dataset of draft-instruction-target triplets. VideoCoCo improves the OmniWeaving baseline from 0.475 to 0.558 on PhyGenBench and from 52.18 to 77.88 on VBench-2.0, achieving the best average score on both benchmarks. These results demonstrate that executable code provides an effective, controllable, and inspectable intermediate representation for physically consistent video generation.

cs.CV↗
arXiv2026

SCAIL-2: Unifying Controlled Character Animation with End-to-End In-Context Conditioning

Controlled character animation aims to transfer motion from a driving sequence to a reference character. Prior works heavily rely on intermediate representations, such as pose skeletons for motion and masked backgrounds for environment, inevitably resulting in information loss. In this work, we present SCAIL-2, a framework that adopts an end-to-end driving paradigm by directly concatenating latent visual information to the model's input sequence. We enable end-to-end training through a data synthesis pipeline that produces MotionPair-60K, a curated dataset for several character animation subtasks. We unify the subtasks using decoupled conditions to accommodate different driving patterns, facilitated by In-Context Mask Conditioning and Mode-Specific RoPE, which provide soft guidance beyond textual instructions and visual information. To address synthetic discrepancy in detailed regions, we propose Bias-Aware DPO to construct preference items to mitigate the errors. Extensive experiments demonstrate that our method achieves state-of-the-art performance across various character animation tasks. Code, model weights, and a large subset of the dataset are available at: https://teal024.github.io/SCAIL-2/.

cs.CV↗
arXiv2026

Auto-AEG: Scalable Data Construction for Open-Vocabulary Audio Event Grounding

Large Audio-Language Models (LALMs) reason fluently about sound yet struggle to localize precisely when events occur, while classical Sound Event Detection attains frame-level precision only over a closed label set. At the intersection of these paradigms lies the task of Open-Vocabulary Audio Event Grounding: predicting all time intervals of a target sound event described by an arbitrary natural language query. Progress is bottlenecked by data scarcity: no large-scale resource provides open-vocabulary onset/offset supervision, and manual temporal annotation is prohibitively expensive. To address this, we introduce Auto-AEG, a scalable pipeline that constructs such supervision by automatic data construction and model fine-tuning. It pairs programmatically synthesized clips, which carry placement-exact ground-truth intervals for supervised cold-start, with multi-model pseudo-labels on real-world audio that supply the reward signal for reinforcement learning. Training with this pipeline yields large temporal-localization gains (+73.9% and +23.1% mIoU over zero-shot) on AEGBench, an independent difficulty-stratified benchmark we release, and these gains generalize to held-out SED and other audio grounding benchmarks. Our results show that automatically constructed data, coupled with interval-aware reward design, provides an effective data-side route to expanding the temporal localization capability of LALMs. AEGBench: https://huggingface.co/datasets/zihan-audio/AEGBench

cs.SD↗
arXiv2026

SCAIL: Towards Studio-Grade Character Animation via In-Context Learning of 3D-Consistent Pose Representations

Achieving controllable character animation that meets studio-grade standards remains challenging despite recent progress. Existing approaches can transfer motion from a driving video to a reference image, but often fail to preserve structural fidelity and temporal consistency in wild scenarios involving complex motion and cross-identity animations. In this work, we present \textbf{SCAIL} (a framework toward \textbf{S}tudio-grade \textbf{C}haracter \textbf{A}nimation via \textbf{I}n-context \textbf{L}earning), which is designed to address these challenges from two key innovations. First, we propose a novel 3D pose representation, providing a robust and flexible motion signal. Second, we introduce a full-context pose injection mechanism within a diffusion-transformer, enabling effective spatio-temporal reasoning over full motion sequences. To align with studio-grade requirements, we develop a curated data pipeline ensuring both diversity and quality, and establish a comprehensive benchmark for systematic evaluation. Experiments show that \textbf{SCAIL} achieves state-of-the-art performance and advances character animation toward studio-grade controlling. Code and model are available at \href{https://github.com/zai-org/SCAIL}{zai-org/SCAIL}.

cs.CV↗
arXiv2026

ProvAgent: Threat Detection Based on Identity-Behavior Binding and Multi-Agent Collaborative Attack Investigation

Advanced Persistent Threats (APTs) pose critical challenges to modern cybersecurity due to their multi-stage and stealthy nature. While provenance-based detection approaches show promise in capturing causal attack semantics, current threat provenance practices face two paradoxical issues: (1) expert skepticism, where human analysts doubt the capability of traditional detection models to identify complex attacks; and (2) expert dependence, as analysts cannot manually process large-scale raw logs to detect threats without these models. Consequently, collaboration between humans and traditional models remains the prevailing paradigm. However, this renders investigation quality contingent upon human expertise and frequently results in alert fatigue. To address these challenges, we present ProvAgent, a framework that evolves the threat provenance paradigm from human-model collaboration to a novel collaboration between multi-agent systems and traditional models. ProvAgent leverages the speed and cost-efficiency of traditional models for initial anomaly screening over large-scale logs. By enforcing fine-grained identity-behavior consistency via graph contrastive learning, it profiles entities based on specific attributes to generate high-fidelity alerts. With these alerts serving as investigation entry points, ProvAgent achieves in-depth autonomous investigation through a hypothesis-verification multi-agent framework. Evaluations with real-world datasets demonstrate that ProvAgent outperforms six state-of-the-art (SOTA) baselines in anomaly detection. Through automated investigation, ProvAgent reconstructs near-complete attack processes at a minimum cost of \$0.06 per day.

cs.CR↗
arXiv2026

Sentient: Detecting APTs Via Capturing Indirect Dependencies and Behavioral Logic

Advanced Persistent Threats (APTs) are difficult to detect due to their complexity and stealthiness. To mitigate such attacks, many approaches model entities and their relationship using provenance graphs to detect the stealthy and persistent characteristics of APTs. However, existing detection methods suffer from the flaws of missing indirect dependencies, noisy complex scenarios, and missing behavioral logical associations, which make it difficult to detect complex scenarios and effectively identify stealthy threats. In this paper, we propose Sentient, an APT detection method that combines pre-training and intent analysis. It employs a graph transformer to learn structural and semantic information from provenance graphs to avoid missing indirect dependencies. We mitigate scenario noise by combining global and local information. Additionally, we design an Intent Analysis Module (IAM) to associate logical relationships between behaviors. Sentient is trained solely on easily obtainable benign data to detect malicious behaviors that deviate from benign behavioral patterns. We evaluated Sentient on three widely-used datasets covering real-world attacks and simulated attacks. Notably, compared to six state-of-the-art methods, Sentient achieved an average reduction of 44% in false positive rate(FPR) for detection.

cs.CR↗
arXiv2025

Entropy-Constrained Strategy Optimization in Urban Floods: A Multi-Agent Framework with LLM and Knowledge Graph Integration

In recent years, the increasing frequency of extreme urban rainfall events has posed significant challenges to emergency scheduling systems. Urban flooding often leads to severe traffic congestion and service disruptions, threatening public safety and mobility. However, effective decision making remains hindered by three key challenges: (1) managing trade-offs among competing goals (e.g., traffic flow, task completion, and risk mitigation) requires dynamic, context-aware strategies; (2) rapidly evolving environmental conditions render static rules inadequate; and (3) LLM-generated strategies frequently suffer from semantic instability and execution inconsistency. Existing methods fail to align perception, global optimization, and multi-agent coordination within a unified framework. To tackle these challenges, we introduce H-J, a hierarchical multi-agent framework that integrates knowledge-guided prompting, entropy-constrained generation, and feedback-driven optimization. The framework establishes a closed-loop pipeline spanning from multi-source perception to strategic execution and continuous refinement. We evaluate H-J on real-world urban topology and rainfall data under three representative conditions: extreme rainfall, intermittent bursts, and daily light rain. Experiments show that H-J outperforms rule-based and reinforcement-learning baselines in traffic smoothness, task success rate, and system robustness. These findings highlight the promise of uncertainty-aware, knowledge-constrained LLM-based approaches for enhancing resilience in urban flood response.

cs.AI↗
arXiv2025

Images are Worth Variable Length of Representations

Most existing vision encoders map images into a fixed-length sequence of tokens, overlooking the fact that different images contain varying amounts of information. For example, a visually complex image (e.g., a cluttered room) inherently carries more information and thus deserves more tokens than a simple image (e.g., a blank wall). To address this inefficiency, we propose DOVE, a dynamic vision encoder that produces a variable number of visual tokens (i.e., continuous representation vectors) to reconstruct each image. Our results show that DOVE significantly reduces the average number of tokens while maintaining high reconstruction quality. In several linear probing and downstream multimodal tasks, it outperforms existing autoencoder-based tokenization methods when using far fewer tokens, capturing more expressive semantic features compared to fixed-length encoding. We further extend DOVE with query-conditioned tokenization. By guiding the model to focus on query-relevant regions, it achieves more efficient and targeted semantic extraction. Our code and checkpoints are available at https://dove-encoder.github.io/dove-encoder.

cs.CV↗
arXiv2024

Winemaking: Extracting Essential Insights for Efficient Threat Detection in Audit Logs

Advanced Persistent Threats (APTs) are continuously evolving, leveraging their stealthiness and persistence to put increasing pressure on current provenance-based Intrusion Detection Systems (IDS). This evolution exposes several critical issues: (1) The dense interaction between malicious and benign nodes within provenance graphs introduces neighbor noise, hindering effective detection; (2) The complex prediction mechanisms of existing APTs detection models lead to the insufficient utilization of prior knowledge embedded in the data; (3) The high computational cost makes detection impractical. To address these challenges, we propose Winemaking, a lightweight threat detection system built on a knowledge distillation framework, capable of node-level detection within audit log provenance graphs. Specifically, Winemaking applies graph Laplacian regularization to reduce neighbor noise, obtaining smoothed and denoised graph signals. Subsequently, Winemaking employs a teacher model based on GNNs to extract knowledge, which is then distilled into a lightweight student model. The student model is designed as a trainable combination of a feature transformation module and a personalized PageRank random walk label propagation module, with the former capturing feature knowledge and the latter learning label and structural knowledge. After distillation, the student model benefits from the knowledge of the teacher model to perform precise threat detection. We evaluate Winemaking through extensive experiments on three public datasets and compare its performance against several state-of-the-art IDS solutions. The results demonstrate that Winemaking achieves outstanding detection accuracy across all scenarios and the detection time is 1.4 to 5.2 times faster than the current state-of-the-art methods.

cs.CR↗
arXiv2023

An Efficient Imbalance-Aware Federated Learning Approach for Wearable Healthcare with Autoregressive Ratio Observation

Widely available healthcare services are now getting popular because of advancements in wearable sensing techniques and mobile edge computing. People's health information is collected by edge devices such as smartphones and wearable bands for further analysis on servers, then send back suggestions and alerts for abnormal conditions. The recent emergence of federated learning allows users to train private data on local devices while updating models collaboratively. However, the heterogeneous distribution of the health condition data may lead to significant risks to model performance due to class imbalance. Meanwhile, as FL training is powered by sharing gradients only with the server, training data is almost inaccessible. The conventional solutions to class imbalance do not work for federated learning. In this work, we propose a new federated learning framework FedImT, dedicated to addressing the challenges of class imbalance in federated learning scenarios. FedImT contains an online scheme that can estimate the data composition during each round of aggregation, then introduces a self-attenuating iterative equivalent to track variations of multiple estimations and promptly tweak the balance of the loss computing for minority classes. Experiments demonstrate the effectiveness of FedImT in solving the imbalance problem without extra energy consumption and avoiding privacy risks.

cs.LG↗