报告人：吴静 博士后研究员 蒙纳士大学

主持人：方鹏飞

报告时间：2025年7月18日（周五）上午9:30

报告地点：av电影 九龙湖校区计算机楼236室

报告摘要：As data privacy regulations (like GDPR and CCPA) become stricter and users grow more aware of their rights, it is increasingly important for machine learning models to be able to “forget” specific data when needed. Machine Unlearning (MU) aims to effectively remove the influence of particular data from a trained model so that the model behaves as if it has never seen the data. This process is crucial not only for protecting user privacy but also for enhancing key attributes of models such as security, fairness, and interpretability. This talk introduces the basics of MU, highlights its challenges, and reviews current mainstream methods. To address the limitations related to gradient conflict resolution and scalability, the talk introduces two key strategies aimed at achieving a better forget-retain trade-off. (1) Prune first, then relearn: identify and reset key parameters to erase the effect of the forgetting data. (2) Game-theoretic optimization: use gradient cooperation between forgetting and retaining objectives, applying Nash bargaining to find a balanced solution. The talk concludes with a discussion of open problems in MU, such as verification, robustness, and data dependency, and outlines future directions, including MU for large language models.

报告人简介：Jing Wu is currently a postdoctoral research fellow in the Department of Data Science & AI at the Faculty of Information Technology, Monash University, Australia. Previously, she served as an assistant lecturer in the Faculty of Engineering at Monash University, where she also completed her PhD in 2025. Her research mainly focuses on Trustworthy ML and Responsible AI, with an emphasis on analyzing the vulnerabilities of deep learning models and developing algorithms to enhance the security and reliability of AI systems, including defenses against attacks and mitigation of undesirable effects. She has published 8 papers in top-tier conferences such as CVPR, ICCV, ECCV, and AAAI. She was recognized as an Outstanding Reviewer at NeurIPS 2024 and a Top Reviewer at ACM MM 2024. Her current and past research services include being the program committees and as a reviewer for several journals and conferences, including IEEE TPAMI, TMLR, CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, AAAI, ECAI, and ACM MM.