Reproduce “Diffusion Sensitivity Geometry and Strong Duality for Optimal Privacy-Utility Tradeoffs” 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
Tue Do, Daniel Alabi
The paper studies privacy for releasing an ideal diffusion denoiser, a posterior mean of training records. It characterizes sensitivity geometrically and chooses Gaussian noise covariance to minimize mean-squared error under a privacy budget. In simulations on [0,1]^10, the naive bound was 1.45 times larger on average and up to 2.5 times in cases; the analysis does not cover learned denoisers.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.