Reproduce “FedSC: Federated Learning with Semantic-Aware Collaboration” 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

Huan Wang, Haoran Li, Huaming Chen, Jun Yan, Jiahua Shi, Jun Shen

FedSC introduces a novel Federated Learning framework that addresses data heterogeneity by leveraging semantic-aware collaboration through a sophisticated prototype-wise approach. The method demonstrates superior performance and faster convergence across diverse non-IID scenarios, including label shift and long-tailed distributions.

Key insights

  • Class-wise prototypes from individual clients tend to be biased due to non-IID data, and naive aggregation of these prototypes at the server dilutes semantic diversity.
  • Merging class-wise prototypes from semantically neighboring clients (Relational Prototypes) enriches local semantic representations and promotes generalizable features.
  • Discrepancy-aware weighted aggregation of relational prototypes (Consistent Prototypes) provides a stable and unbiased global semantic signal, especially for imbalanced data.

Summary by alphaXiv.

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