Advancing Machine Unlearning in Deep Models: Causal Insights, Efficiency, and Effectiveness

The School of EECS is hosting the following Seminar:

Advancing Machine Unlearning in Deep Models: Causal Insights, Efficiency, and Effectiveness

Speaker: Shaofei Shen
Host: Prof Hongzhi Yin

Abstract: In the current context of heightened data privacy awareness, the concept of machine unlearning has gained increased attention. This process aims to remove the private information derived from forgotten data in a well-trained model while maintaining the integrity of the remaining dataset. To handle the unlearning requests, the unlearning algorithm requires removing the knowledge of forgetting data with the following three objectives: 1) removing forgetting data cannot affect the normal functionality of the remaining data; 2) increasing the data and time efficiency for handling the unlearning request; 3) accurately processing unlearning requests to ensure the information to be removed. Nonetheless, previous methods often encounter obstacles that prevent the realization of unlearning objectives: difficulty maintaining performance on remaining data, low data and time efficiency, and challenges in evaluating the unlearning performance. To address such issues, this PhD research explores three key aspects: 1) exploring the causes of performance degradation and maintaining the performance on the remaining data and test data., 2) reducing the reliance on label information and improving the time cost during the unlearning procedure, and 3) introducing a layer-wise and instance-wise evaluation for the unlearning algorithm and results. The overall aim of this research is to advance the field of machine unlearning in the aspects of causal insights, efficiency, and effectiveness.

Biography: Mr Shaofei Shen received a B.Sc. degree in mathematics and applied mathematics from the University of Science and Technology of China and a M.Sc. degree in data science from the University of Queensland. He is currently pursuing a Ph.D degree at the School of EECS under the supervision of Dr Miao Xu, Dr Alina Bialkowski, and Dr Weitong Tony Chen. His research interests encompass machine learning and unlearning.