一、报告主题：PPO-TVC: A Trust-Aware Deep Reinforcement Learning Framework for Edge Video Caching against Sybil Attacks

二、报告人：李光辉

三、报告时间：2026年10月21日（周三） 15：30

四、报告地址：计算机大楼B503

五、报告摘要：

The explosive growth of short-video platforms has shifted network traffic dominance toward user-generated content, making edge caching pivotal for alleviating backhaul congestion. However, accurately estimating video popularity is challenging due to the prevalence of Sybil attacks, which manipulate engagement signals and cause severe cache pollution. To address this issue, we propose PPO-based trust-aware video caching (PPO-TVC), a deep reinforcement learning framework designed to ensure robust caching performance under adversarial traffic. Our approach leverages a hybrid trust modeling module that combines clustering with graph-based propagation to extract credibility-aware user signals. By incorporating trust-weighted value estimates into the learning agent, we effectively filter out Sybil-induced noise. Furthermore, we introduce a decoupled actor-critic architecture that decomposes per-video value estimation. This design effectively overcomes the curse of dimensionality in large action spaces, enabling scalable, real-time decision-making. Extensive experiments on the real-world KuaiRand dataset demonstrate that PPO-TVC significantly outperforms representative baselines. It achieves superior cache hit rates and lower latency while maintaining high resilience against varying intensities of Sybil attacks.

六、报告人简介：

李光辉，江南大学人工智能与计算机学院教授、博士生导师，曾入选浙江省“新世纪151人才工程”第二层次、杭州市钱江特聘专家、江苏省无锡市社会事业领军人才。主要从事物联网、边缘计算、智能无损检测等领域的研究和教学工作。主持承担过国家863计划项目1项、国家自然科学基金重大研究计划培育项目1项和面上项目6项，国防科技创新特区课题1项，以及江苏省重点研发计划等省部级科研项目15项。在IEEE/ACM TMC、TKDE、TKDD、TSC、TITS、TGRS、计算机学报、软件学报等国内外期刊或学术会议发表论文150余篇。作为第一完成人，先后获得浙江省科技进步二等奖和江苏省科学技术奖二等奖。现任中国计算机学会物联网专委、容错计算专委和普适计算专委的执行委员，以及江苏省计算机学会计算机与通信专委会副主任。