Reproduce “MOMENT-DP: Momentum Aware Evaluation of Correlated Noise for Differentially Private Training” 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
Sunny Gupta, Raja Viswanadha
MOMENT-DP evaluates how privacy-valid correlated-noise mechanisms affect training when consumed by optimizers with persistent state. It models how momentum, curvature, and the learning-rate schedule shape noise’s contribution to final parameter error, extending a quadratic criterion that matches NoiseCurve when momentum is zero. In controlled tests, the best mechanism changes with momentum; logistic regression supports the pattern, but neural transfer remains unresolved.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.