Reproduce “Class-Wise Federated Averaging for Efficient Personalization” 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
Gyuejeong Lee, Daeyoung Choi
cwFedAvg personalizes federated learning by aggregating client models into class-specific global models and combining them according to each client’s class mix. Weight Distribution Regularizer (WDR) encourages output-layer weight norms to estimate class proportions without sending class counts. On practical CIFAR-100, output-layer-only cwFedAvg reaches 56.29 ± 0.18% accuracy versus 47.65 ± 0.62% for FedAMP. This is a benchmark result; the server stores multiple global models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.