Title: Protecting the Undeleted in Machine Unlearning
Speaker：Prof. Kobbi Nissim (Georgetown University)
Date and time: June 4 at 2:00 PM.
Online Venue (Zoom): The URL will be provided only to registered participants.
Abstract: Legal data protection standards such as the EU General Data Protection Regulation and the California Consumer Privacy Act give individuals the right to request that their specific information be deleted, also known as the Right to be Forgotten. This provision gave rise to machine unlearning, a branch of machine learning focused on removing elements from training data by efficiently producing a model that would have been obtained had the deleted data never been included, namely, “perfect retraining.”
In this talk, Prof. Nissim will discuss how data deletion affects privacy. He will first present a task that can be computed with strong privacy guarantees, yet any perfect retraining mechanism for the task allows an adversary controlling only a small number of data points to reconstruct almost the entire dataset simply by issuing deletion requests.
He will then discuss ways forward, in particular a new cryptographically motivated security definition that safeguards undeleted data points against leakage caused by the deletion of other points. The talk will also show that this definition permits several essential functionalities, including bulletin boards, summations, and statistical learning.
This is based on joint work with Aloni Cohen, Refael Kohen, and Uri Stemmer.
Bio: Prof. Kobbi Nissim is McDevit Chair in Computer Science at Georgetown University and is affiliated with Georgetown Law. His work focuses on the mathematical formulation and understanding of privacy. His work with Dinur and Dwork in 2003 and 2004 initiated rigorous foundational research on privacy, and in 2006 he introduced differential privacy with Dwork, McSherry, and Smith. Prof. Nissim is a Fellow of the IACR. He received the Paris Kanellakis Theory and Practice Award in 2020, the Caspar Bowden Award for Outstanding Research in Privacy Enhancing Technologies in 2019, the Gödel Prize in 2017, and test-of-time awards in 2013 and 2018.
