Reproduce “Revisiting Gradient Pruning: A Dual Realization for Defending against Gradient Attacks” 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
Lulu Xue, Shengshan Hu, Ruizhi Zhao, Leo Yu Zhang, Shengqing Hu, Lichao Sun, Dezhong Yao
DGP reduces gradient-inversion risk and communication cost in collaborative learning by pruning both the largest and smallest gradient entries, with error feedback. In CIFAR10 ResNet18 tests at 80% pruning, it yields less similar reconstructions than Top-k under the passive IG attack. For ResNet18, communication per epoch falls from 85.251 MB to 52.216 MB; these results do not establish general privacy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.