Reproduce “Test-Time Robust Personalization for Federated Learning” 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

Liangze Jiang, Tao Lin

FedTHE addresses changing test distributions in federated learning by combining a global classifier with a client-personalized classifier. At deployment, it uses each unlabeled example to adapt their mixture, and FedTHE+ additionally performs unsupervised test-time fine-tuning. On CIFAR10 with Dirichlet(0.1) clients, FedTHE+ reaches 58.29% out-of-client accuracy, below FedAvg’s 64.31%, while its local ID accuracy is 90.55%.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.