WLWeiguo Lin

Weiguo Lin

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Weiguo Lin has a total of 63 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    From Design Choices to Engineering Risks:A Large-Scale Empirical Study of Best Practices and Defectsin Developer-Facing Open-Source Agent Frameworks

    by Yibo Zhai, Shenao Wang, Junjun Si, Yanjie Zhao, Weiguo Lin, Haoyu Wang

    The rapid development of Large Language Models (LLMs) has accelerated the adoption of LLM agents in software engineering, data analysis, information retrieval, and autonomous tasks. Popular open-source frameworks such as LangChain have become important development kits for building LLM agents. However, existing research has mainly examined LLM agents from the perspectives of safety, security, and performance, leaving limited empirical evidence on how the underlying frameworks shape agent construction practices, downstream adoption, and quality risks. To address this gap, we present a large-scale empirical study of practices and defects in open-source LLM agent projects. We collected 116 candidate agent frameworks, identified 10 popular frameworks, and characterized them through popularity analysis, repository maintenance, and lifecycle analysis. We synthesize 14 agent-construction best practices from official documentation, API references, and public guidance, and operationalize seven into 10 observable construction-defect patterns. We then compare framework support and analyze 15,042 framework-project observations from 73,014 repositories, corresponding to 13,002 deduplicated reported downstream projects. Results show uneven framework support, with LangGraph achieving the highest aggregate score, followed by LangChain and CrewAI. Downstream defects are widespread: 96.79\% of reported projects contain at least one defect, 75.33\% contain at least two, and 4.38\% contain four or more.

  • Open Access

    Engineering Practices in Open-Source Agent Frameworks: A Large-Scale Empirical Study

    by Yibo Zhai, Junjun Si, Yan Wang, Weiguo Lin

    The rapid development of Large Language Models (LLMs) accelerates the adoption of LLM-based agents in software engineering, data analysis, and automated decision-making. To support agent development, various open-source frameworks provide modular abstractions for planning, tool invocation, and memory management, lowering the engineering barrier. Although widely used, research mainly emphasizes capability modeling and architectural innovation, while systematic analysis of engineering practices and real-world usage remains limited. This paper presents a large-scale empirical study of mainstream open-source agent frameworks from an engineering perspective. We first identify the ten most popular frameworks in the open-source community and examine their popularity and language support. By analyzing official documentation, API interfaces, and source-level implementations, we extract common design principles and distill five best practices. We then formalize corresponding defect patterns and perform stratified proportional sampling on 73,014 downstream projects to evaluate how these practices are implemented. The results reveal recurring issues, including incomplete agent description, improper memory Strategy, insufficient monitoring and error handling, and suboptimal tool design. These defects appear across frameworks, indicating structural gaps between recommended abstractions and practical adoption. Overall, this study provides empirical evidence linking framework design choices to engineering risks and highlights the need for stronger guidance and support in agent framework development.

  • LaQual: An automated framework for LLM app quality evaluation

    by Yan Wang, Xinyi Hou, Junjun Si, Yanjie Zhao, Weiguo Lin, Haoyu Wang

  • Open Access

    Few-Shot Cross-Domain Deepfake Detection for Edge Devices: A Feature Decoupled System Architecture

    by Zhenpeng Ai, Junfeng Xu, Weiguo Lin

    Deploying highly generalizable deepfake detection systems on resource-constrained edge devices poses a significant technical challenge for conventional end-to-end large models that rely heavily on computational resources. Extracting multi-source physical prior features is a viable approach under limited computational power; however, in few-shot scenarios, the dimensional mismatch of heterogeneous features is prone to causing downstream classifiers to overfit. To mitigate this bottleneck, this paper proposes a “static feature extraction–central normalization alignment–independent downstream decision” decoupled detection system for few-shot cross-domain tasks on edge devices. The front end of the system constructs an 856-dimensional comprehensive feature reservoir, and a lightweight residual normalization adapter gϕ is introduced as the central support module. This module explicitly compresses the intra-class variance of heterogeneous features, providing a smoothly aligned manifold base for downstream classifiers. Experimental results indicate that this decoupled architecture demonstrates consistent stability in few-shot (K=10) cross-domain evaluations. When encountering intra-family cross-domain shifts and cross-mechanism distribution shifts from diffusion models, the accuracy reaches 84.9% and 76.1%, respectively. Compared to representative end-to-end meta-learning baselines (e.g., MAML), the relative error rate is reduced by over 30%. Furthermore, after completing the asynchronous offline pre-processing (approximately 897 ms) at the front end, a single-image online classification query requires only 7.7 ms under a simulated single-core CPU constraint, satisfying the low-latency requirements for lightweight deployment on edge devices. Finally, combined with empirical observations, this paper discusses the performance boundaries of the architecture in cross-mechanism metric mismatch scenarios, providing a low-barrier, robust engineering defense scheme for resource-constrained environments.

  • Adversarial example generation for infrared images

    by Yuefeng Wang, Weiguo Lin, Junfeng Xu, Yikun Xu, Yong Gan


  • Robust and Unstigmatized Imperceptible Perturbations for Rendering Face Manipulation Ineffective

    by Yibo Zhang, Weiguo Lin, Zhihong Tian, Geyong Min, Junfeng Xu, Yikun Xu

    The widespread adoption of face manipulation systems has brought entertainment and convenience to users while posing significant challenges to media forensics. Conventional active defense strategies typically generate adversarial images by introducing perturbations into the original images. When adversarial images undergo facial manipulation, they often exhibit distortions or speckle artifacts, which helps reduce the dissemination of forged content on social media platforms. Nevertheless, the widespread dissemination of degraded images may contribute to facial stigmatization. Furthermore, conventional perturbation techniques are vulnerable to failure under JPEG compression and various image processing operations on OSN platforms. To address these challenges, we introduce a robust and unstigmatized imperceptible perturbation (RUIP) method designed to counteract face manipulation. First, RUIP utilizes an end-to-end adversarial training framework to generate robust and imperceptible perturbations. Second, to mitigate facial stigmatization, we incorporate both pixel-level and feature-level guidance losses during training, ensuring that the output images remain visually natural and closely aligned with the original images. Finally, we develop a novel module, the Flexible Random Enhancement Generator (FREG), to simulate complex JPEG compression and diverse image processing operations on OSN platforms, enhancing the model’s robustness against perturbations. Extensive qualitative and quantitative experiments demonstrate that the proposed method effectively defends against face manipulation attacks while preserving the visual quality of facial images under JPEG compression and other image processing operations on OSN platforms. We propose an effective and unstigmatized defense algorithm to safeguard privacy and maintain the stability of the social media ecosystem. Code is available at https://github.com/silencecmsj/RUIP.


  • Open Access

    Leveraging vision transformers and entropy-based attention for accurate micro-expression recognition

    by Yibo Zhang, Weiguo Lin, Yuanfa Zhang, Junfeng Xu, Yan Xu

    Micro-expressions are difficult to fake and inherently truthful, making micro-expression recognition technology widely applicable across various domains. With the development of artificial intelligence, the accuracy and efficiency of micro-expression recognition systems have been significantly improved. However, the short duration and subtle facial movement changes present significant challenges to real-time recognition and accuracy. To address these issues, this paper proposes a novel micro-expression recognition method based on the Vision Transformer. First, a new model called HTNet with LAPE (hierarchical transformer network with learnable absolute position embedding) is introduced to improve the model’s capacity for capturing subtle facial features, thereby enhancing the accuracy of micro-expression recognition. Second, an entropy-based selection agent attention is proposed to reduce the model parameters and computational effort while preserving its learning capability. Finally, a diffusion model is utilized for data augmentation to expand the micro-expression sample size, further enhancing the model’s generalization, accuracy, and robustness. Extensive experiments conducted on multiple datasets validate the framework’s effectiveness and highlight its potential in real-world applications.