Reproduce “Actually, All Learnable Distribution Classes Are Privately Learnable” 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
Mohammad Afzali, Hassan Ashtiani, Alex Bie, Gautam Kamath, Peihan Liu, Argyris Mouzakis, Vikrant Singhal
The authors show that every distribution class learnable to arbitrarily small total-variation error on standard Borel domains is also learnable with (ε, δ)-differential privacy. They run a non-private learner on data blocks, couple synthetic samples from its estimates so close estimates often match exactly, and privately select a frequent sample. This is a sample-complexity theorem, not an efficiency result.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.