Reproduce “Rethinking Model Aggregation for Federated Learning with Domain Skew: A Knowledge-Preserving Perspective” 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

Haotian Wu, Xu Ji, Qiheng Li, Li Xiong, Zibin Pan

FKPL addresses federated image classification across clients with shared labels but different visual domains by preserving group-trained encoders and training a shared classifier on their concatenated features. Prototype-based classifier calibration (PCC) uses client-uploaded labeled virtual features to train it on the server. With the extractor fixed, Office-Caltech-10 mean accuracy rises from 63.02% to 73.01%, not establishing gains on unseen domains.

Summary by alphaXiv.

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