FedELR: When federated learning meets learning with noisy labels

Ruizhi Pu, Lixing Yu, Shaojie Zhan, Gezheng Xu, Fan Zhou, Charles X. Ling, Boyu Wang

Abstract: Highlights•We study the label noise in federated learning (FL).•We observe mismatches of ETP between client and global models in FL under noisy labels.•We show label noise causes local/global inconsistency that hinders global optima.•We propose a lightweight FL framework using ELR to harness early learning against noise.