Learning About Unlearning: The Promise and Progress of Machine Unlearning

Rapid advancements in artificial intelligence present novel challenges for upholding and enforcing important privacy and intellectual property rights. The growing field of machine unlearning holds promise to help meet these challenges by developing solutions to make a machine learning model efficiently and verifiably “forget” certain data it was trained on. Experts on this cutting-edge area of research will introduce the latest techniques and highlight ongoing efforts to measure success. The session will explore the implications of machine unlearning for data deletion rights, copyright infringement and algorithmic disgorgement remedies.

What you will learn:

• Categories of machine unlearning techniques, including their benefits, drawbacks and current and future use cases
• How machine unlearning features in emerging legal enforcement and policy.
• Current and future research efforts to create scalable and efficient algorithms as well as quantitative evaluation metrics to measure unlearning effectiveness and model utility.