Tag
The paper introduces SWB-DM, a robust aggregation method for federated learning that combines sliced Wasserstein barycenters with delayed momentum caching to handle Byzantine attacks and partial participation effectively.