loss-based-weighting

This paper proposes FedLBW, a federated learning aggregation strategy that weights client updates by inverse validation loss instead of dataset size, improving accuracy and robustness to non-IID data and client dropouts in wireless networks.