Reproduce “Pre-Pruning and Gradient-Dropping Improve Differentially Private Image Classification” 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
Kamil Adamczewski, Yingchen He, Mijung Park
DP-SSGD trains private image classifiers by pruning connections before training and selecting a subset of remaining weights for each private update. It reports 57.45% top-1 accuracy on from-scratch CIFAR-10, versus 56.8% for the De et al. baseline at ε=1, a difference of 0.65 percentage points across five runs. The result does not establish statistical significance or a general speedup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.