Machine Unlearning
Description
This book is a comprehensive guide to machine unlearning, covering both theoretical foundations and practical algorithms. The first part develops data influence measurement methods, including real-time and time-varying valuation frameworks. The second part presents exact and approximate unlearning approaches for large-scale models, with a focus on wireless and networked systems.
As AI models face growing demands to remove specific training data due to privacy regulations, security threats, or data quality concerns, machine unlearning has emerged as an efficient alternative to costly full retraining. This challenge is particularly critical in networked environments where user-generated data is continuously produced at scale.
This book is designed for researchers and graduate students in computer science, AI, and data privacy who seek to understand machine unlearning and explore open research challenges. It is also useful to industry practitioners in telecommunications and edge computing who need practical solutions for data removal and privacy compliance. By covering both current methods and future directions such as federated unlearning and unlearning for foundation models, this book provides a clear roadmap for advancing machine unlearning and building more trustworthy and adaptable AI systems.
More details
Persons
Jie Xu is currently a Postdoctoral Fellow in the Department of Computer Science at City University of Hong Kong. She received her B.Eng. degree in Information Security and B.A. degree in Communication from the University of Science and Technology of China (USTC) in 2017, M.Eng. degree in Electronics and Communication Engineering from USTC in 2020, and Ph.D. degree in Computer Science from City University of Hong Kong in 2024. She is a recipient of the CityU Presidential Ph.D. Scholarship and the Best Paper Runner-Up Award at IEEE MASS 2018. Her research interests include trustworthy artificial intelligence, distributed systems, and data privacy. Her work has appeared in leading venues including ICML, ACL, and ICLR.
Xiaohua Jia is an IEEE Fellow and ACM Fellow. He is currently a Chair Professor in the Department of Computer Science at City University of Hong Kong and Director of the Center of Decentralized Trust Computing (CDTC). He received his BSc and MSc in Computer Science from the University of Science and Technology of China in 1984 and 1986, respectively, and his DSc degree in Information Science from the University of Tokyo in 1991. His research interests include distributed systems, data privacy and security, and cloud computing. He serves as an Editor for IEEE Transactions on Computers and has chaired major conferences including IEEE ICDCS 2023, ACM ICN 2019 and ACM MobiHoc 2008.
Content
.- Introduction.
.- Background of Machine Unlearning.
.- Real-Time Data Influence Measurement.
.- Time-Varying Data Influence Measurement.
.- Exact Unlearning for Large Models via Prompt Tuning.
.- In-Training Unlearning for Large Models.
.- Conclusion, Practical Challenges, and Future Directions.