Reproduce “Privately Clipping Heavy-Tailed Data” 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 Joseph, Alex Kulesza
OptimizeRC privately selects a clipping bound for private vector sums using residual coherence, the summed unit directions of vectors above a candidate bound. Unlike norm quantiles, it captures tail alignment to balance clipping bias and noise. Across ten heavy-tailed datasets, it ranked best on six and second-best on four, with a worst error ratio of 1.32; this benchmark does not prove near-optimality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.