Watch Your Head: Assembling Projection Heads to Save the Reliability of Federated Models

arXiv:2402.16255v1 Announce Type: new
Abstract: Federated learning encounters substantial challenges with heterogeneous data, leading to performance degradation and convergence issues. While considerable progress has been achieved in mitigating such an impact, the reliability aspect of federated models has been largely disregarded. In this study, we conduct extensive experiments to investigate the reliability of both generic and personalized federated models. Our exploration uncovers a significant finding: \textbf{federated models exhibit unreliability when faced with heterogeneous data}, demonstrating poor calibration on in-distribution test …

abstract arxiv challenges convergence cs.ai cs.lg data federated learning head impact performance progress projection reliability save study type