FEBRUARY MEETING: Thursday, February 26, 2026 at 5:30pm (US Central)
NEW LOCATION:
ESAM Conference Room, Tech M416
2145 Sheridan Road, Evanston, IL 60208
AGENDA:
5:30pm - Meet and greet with refreshments
6:00pm - Talk with Siqiao Mu, Ph.D. Candidate, Department of Engineering Sciences and Applied Mathematics, Northwestern University
TALK TITLE:
Gradient Algorithms for Machine Unlearning
ABSTRACT:
Machine unlearning algorithms aim to efficiently remove data from a model without retraining it from scratch, in order to remove corrupted or outdated data or respect a user's "right to be forgotten." Since empirical unlearning heuristics can be unreliable, we desire "certified" machine unlearning algorithms, which are theoretically guaranteed to achieve probabilistic indistinguishability between the unlearned model and the model retrained on the retained data samples. While several works have proposed second-order unlearning algorithms, first-order methods such as gradient descent (GD) or stochastic gradient descent (SGD) algorithms are far more computationally tractable for large neural networks. We propose "Rewind-to-Delete," a first-order unlearning algorithm that is also black-box, in that it can be applied to trained models without costly precomputation. We prove certified unlearning guarantees and derive privacy-utility-complexity tradeoffs for both the GD and SGD versions.
DATA SCIENCE NIGHTS are monthly meetings featuring presentations and discussions about data-driven science and complex systems, organized by Northwestern University graduate students and scholars. Students and researchers of all levels are welcome! For more information: http://bit.ly/nico-dsn
FUTURE DATES:
Data Science Nights will be held on Thursday evenings in the winter and spring terms, with future dates on March 19, April 30, and May 28, 2026.