Reproduce “Remember What You Want to Forget: Algorithms for Machine Unlearning” 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
Ayush Sekhari, Jayadev Acharya, Gautam Kamath, Ananda Theertha Suresh
The authors present a data-unlearning method that uses requested examples and a Hessian statistic saved during training to update a model and add calibrated Gaussian noise. Deletion capacity is the largest number of requests maintaining excess population risk at most 0.01. For convex losses, its dimension dependence is d^-1/4, versus d^-1/2 for the compared DP-style method; this is theoretical, not universal.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.