Near-Optimal Machine Unlearning Utility for Smooth Strongly Convex Losses
This work nearly resolves the fundamental statistical cost of unlearning by proving upper and lower bounds on the excess population risk of approximate $(\varepsilon, \delta)$-unlearning; the authors' bounds are tight up to a condition-number factor.
Matthew Regehr, Gautam Kamath, Andrew Lowy
· 1 citation