Reproduce “A Novel Federated Learning Framework for Defending Against Gradient Inversion 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
Pengxin Guo, Ziying Huang, Shuang Zeng, Xiaodan Zhang, Weihong Ren, Yuyin Zhou, Liangqiong Qu
HyperFL++ changes federated learning by having clients share compressed hypernetwork parameters rather than ordinary model updates, while keeping classifiers and client embeddings local. Row-wise Kronecker Decomposition reduces the shared parameter count. In CIFAR-10 gradient-inversion attack tests, its reconstructions scored PSNR 8.31, SSIM 0.2796, and LPIPS 0.7657, versus FedAvg’s 16.16, 0.6415, and 0.0536; this is one attack test, not a general privacy guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.