ML Theory Seminar: A New Approach to Machine Unlearning
Date:
The next ML Theory Seminar is scheduled for March 18 at 11:00 AM in Sage 3713, RPI, and will be presented by Sadia Asif.
Sadia will present her accepted ICLR 2026 work on machine unlearning. The talk introduces an approach derived from the optimality conditions of a bi-level unlearning formulation, rather than the standard scalarized objective that balances forgetting and retention. The resulting algorithm provides convergence-rate guarantees and demonstrates a stronger trade-off between targeted forgetting and retained model utility compared with prior methods.