Nonlinear Equilibrium Transitions in a Potential Game Model for Federated Learning
Researchers have developed a potential game framework to model federated learning scenarios where clients act out of self-interest. This model analyzes how clients' rational choices in training efforts, influenced by server rewards, lead to Nash equilibria. The study reveals that these equilibria transition nonlinearly with a critical reward factor, potentially causing a shift between low and high effort levels among clients. The paper also validates the effectiveness of this critical factor in federated learning training. AI