Machine Unlearning: Foundations, Algorithms, and Advances in Quantum Technology
Description
This book explores the cutting-edge concept of machine unlearning and its application across various fields, especially within AI and machine learning models. It addresses the critical need to "forget" specific data in models to comply with evolving privacy regulations, enhance model robustness, and mitigate security risks. With a focus on real-world implications, this book presents a thorough analysis of unlearning techniques and frameworks, detailing approaches from exact data removal to approximate, efficient methods that support high-performance models in dynamic environments.
The chapters delve into machine unlearning for large language models, addressing privacy concerns in unstructured data, and the challenges of catastrophic recalling. Each chapter provides readers with actionable insights into the mechanisms, benefits, and trade-offs involved in implementing unlearning. Readers will discover pioneering frameworks, such as federated fuzzy unlearning, and advanced techniques that combat over-unlearning, ensuring model integrity without extensive retraining.
This book is designed for researchers, AI practitioners, and data scientists interested in integrating unlearning for ethical, secure, and adaptive AI systems. A foundational knowledge in AI or machine learning is recommended. By the end, readers will gain a robust understanding of unlearning methodologies and practical strategies to implement them within various applications, driving responsible AI innovation.
More details
Persons
Youyang Qu is a Professor at Qilu University of Technology (Shandong Academy of Sciences), who is specializing in data privacy, machine learning, and federated learning systems. His work has notably contributed to the field of machine unlearning, where he has authored five key publications that explore unlearning techniques in large language models, federated learning frameworks, and privacy-preserving methods for distributed AI. He has a robust background in both theoretical and applied AI research, combining advanced data governance principles with practical solutions. His contributions to the field have helped advance unlearning as a crucial area in privacy-aware machine learning.
Xiaoming Wu is a professor at Qilu University of Technology (Shandong Academy of Sciences) in Jinan, China. He received his Ph.D. in software engineering from the Shandong University of Science and Technology in 2017. His research interests include high-performance computing, computer networks, and information security. He has contributed to various publications in these fields and has been involved in organizing international conferences, such as serving as a General Chair for 2024 IEEE International Conference on Data Science in Cyberspace.
Guobin Zhang is an Associate Professor at the School of Law, Shanghai Jiao Tong University, specializing in maritime law and international seabed legislation. His analysis of China's 2016 seabed legislation and regulatory frameworks demonstrates expertise applicable to data governance. Zhang's methodologies in analyzing complex legal systems and international compliance inform robust data governance policies. His insights into treaty applications and cross-border regulations equip him to address data protection standards and formulate strategies for global data management. His expertise makes him a valuable voice in ensuring data security, privacy, and ethical governance.
Shaoting Tang is a full professor at Beihang University, specializing in complex systems, nonlinear dynamics, network science, brain science, artificial intelligence, and evolutionary dynamics. Her research encompasses foundational advancements in swarm intelligence, opinion dynamics, information diffusion, and brain network analysis. Collaborating with leading experts worldwide, Professor Tang has developed models that incorporate confirmation bias and peer pressure in opinion dynamics, providing insights into public opinion evolution. Her extensive publication record includes articles in top-tier journals such as Physicis Review X, Nature Communication and PNAS, significantly influencing both academic research and practical applications in understanding swarm intelligence, complex systems, and human behavior.
Longxiang Gao is a Professor Qilu University of Technology (Shandong Academy of Sciences) with expertise in distributed AI, federated learning, and privacy-centric machine learning applications. His research, often in collaboration with Youyang Qu, includes pioneering studies on unlearning in multi-task federated systems and privacy-preserving swarm intelligence. Co-authoring three of the recent papers in this domain, Gao has provided critical insights into using fuzzy logic for federated unlearning and mitigating over-unlearning in AI models. His work continues to impact the development of resilient, privacy-focused AI solutions, particularly in consumer electronics and decentralized learning environments.
Phillip S. Yu is a Distinguished Professor at the University of Illinois Chicago and a world-renowned expert in big data, data mining, and privacy-preserving machine learning. His research spans foundational advancements in distributed computing, federated learning, and scalable AI systems.
Content
Chapter 1. Introduction.- Chapter 2: The Foundations of Machine Unlearning.- Chapter 3: Machine Unlearning in Large Language Models (LLMs).- Chapter 4: Continuous Unlearning Verification.- Chapter 5: Mitigating Over-Unlearning.- Chapter 6: Unlearning in Distributed Scenarios.- Chapter 7: Summary and Future Directions.