SERC Symposium: Speaker Bio and Abstract
Ashia Wilson, Assistant Professor of Electrical Engineering and Computer Science, MIT
Complying with privacy and copyright requests
The proliferation of machine learning tools including large language model has led to mounting privacy and copyright concerns. With performance as the priority, many companies are reluctant to train their models in a differentially private way, resulting in the need for post hoc tools to comply with any data removal requests that might arise. The area of “machine unlearning” focuses on developing such tools from removing data points from the model in a seamless and efficient way. This talk reviews several of the proposed techniques for removing data and discusses open questions about whether data can be removed from modern machine learning models efficiently.
Ashia Wilson works in the Department of Electrical Engineering and Computer Science, as an Assistant Professor. Ashia received her PhD in Statistics from the University of California, Berkeley, and her BA in Applied Mathematics from Harvard. Her research centers upon optimization, algorithmic decision making, dynamical systems, and fairness within large scale machine learning.
A National Science Foundation Graduate Research Fellow, Ashia has received the NeurIPS ’17 Spotlight Paper Award for The Marginal Value of Adaptive Methods in Machine Learning and has performed research with Microsoft and Google AI. Her papers have been published in the Proceedings of the National Academy of Science, in Advances in Neural Information Processing Systems, and in the International Conference of Machine Learning, among others. Additionally, she has served as a reviewer for NeurIPS and the Journal of Machine Learning.