Reproduce “Differential Privacy Meets Neural Network Pruning” 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, Mijung Park
DP-SSGD uses public data to prepare models and privately update selected weights through fixed freezing or stepwise selection. By limiting coordinates receiving noisy updates, it may preserve useful signal under differential privacy. On CIFAR-10, ResNet-20 reaches 72.11% test accuracy with random freezing at ε=2, versus 68.5% for unpruned DP-SGD, also publicly pretrained; outcomes depend on privacy setting and public-data compatibility.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.