AI RESEARCH
Federated Learning by Utility-Constrained Stochastic Aggregation for Improving Rational Participation
arXiv CS.LG
•
ArXi:2605.18020v1 Announce Type: new Federated Learning (FL) algorithms implicitly assume that clients passively comply with server-side orchestration by sharing local model updates upon server request. However, this overlooks an important aspect in real-world cross-silo environments: clients are often rational agents who may prioritize their utilities such as local model performance over that of the global model.