Reproduce “Near-Optimal Machine Unlearning Utility for Smooth Strongly Convex Losses” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Matthew Regehr, Gautam Kamath, Andrew Lowy
The authors characterize the minimax excess risk of machine unlearning in smooth, strongly convex stochastic optimization. Their method shifts probability between a concentrated output near the full-data solution and a wider region around the retained-data solution, achieving pure ε-unlearning. For mean estimation, the rate is 1/n+(m/n)^2e^{-2ε/(d+2)}, versus retraining’s 1/n+(m/n)^2; this is a statistical rate, under stated assumptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.