Reproduce “Variational Bayes In Private Settings (VIPS)” 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
Mijung Park, James Foulds, Kamalika Chaudhuri, Max Welling
VIPS makes iterative Bayesian inference differentially private by keeping per-record latent-variable calculations inside a trusted boundary and adding calibrated noise to expected sufficient statistics used for model updates. It combines subsampling with moments-accountant composition. On tested public Wikipedia LDA settings, it outperformed strong composition in the privacy-perplexity trade-off; this finding is limited to those settings and does not establish results on confidential data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.