Prof. Zhou Wanlei, vice rector of City University of Macau (CityU) and dean of the Faculty of Data Science, recently visited Jinan to attend the 9th IEEE International Conference on Data Science in Cyberspace (IEEE DSC 2024) at the invitation. At the conference, Prof. Zhou delivered a keynote speech titled Machine Unlearning: Realising the ‘Right to be Forgotten,' discussing research findings on machine unlearning. Additionally, the research paper Defense Against Graph Injection Attack in Graph Neural Networks, co-authored by Prof. Zhou and Prof. Zhu Tianqing, associate dean of the Faculty of Data Science (FDS), was awarded the Outstanding Paper Award by the conference.
During his presentation, Prof. Zhou Wanlei introduced the background and current development of machine unlearning and proposed several methods for enforced unlearning. He pointed out that machine learning has a positive impact to our daily lives and has made significant progress over the past decade. “Its application covers different areas, such as loan applications, facial recognition, recidivism prediction, and large language models,” said Prof. Zhou, “as such machine unlearning emerged due to concerns regarding privacy, information security, applications, and legal regulations.”
Prof. Zhou Wanlei discussed the mechanisms behind machine unlearning and differentiated it from different concepts such as catastrophic forgetting, data masking, differential privacy, and online learning. He proposed two methods for enforced unlearning: data reorganization and model manipulation. Additionally, he outlined two methods for verifying the successful implementation of machine unlearning: empirical evaluation and theoretical calculation. Lastly, Prof. Zhou introduced practical cases illustrating the application scenarios of machine unlearning and highlighted the challenges facing in the future.
Additionally, Professor Zhou, Professor Zhu Tianqing, and their research team studied attack behaviors in Graph Neural Networks and proposed a defense method against link stealing attacks, demonstrating its effectiveness. Their research findings were compiled into the paper Defense Against Graph Injection Attack in Graph Neural Networks which was presented by Prof. Zhou during the conference. The presentation was very well received and won the Outstanding Paper Award.
The International Conference on Data Science in Cyberspace was organized by IEEE (Institute of Electrical and Electronics Engineers), the IEEE Technical Committee on Scalable Computing, the IEEE Computer Society, the Chinese Academy of Engineering, and the Chinese Information Processing Society of China. It focused on data science and its applications in cyberspace, and has become one of the most influential and attractive international conferences. Over the years, experts, scholars, and industry representatives from both domestic and international backgrounds in the field of cyberspace data science have engaged in extensive exchanges on research frontier, including data science, big data, data applications in cyberspace, social networks and media, data mining, machine learning, and big data security and privacy protection.
Source: 9th IEEE International Conference on Data Science in Cyberspace (IEEE DSC 2024)