Prospect Theory-Based Federated Learning Incentive Mechanism for Industrial IoT

As an emerging technology, federated learning (FL) plays a critical role for information sharing in industrial Internet of Things (IIoT). FL integrates information from multiple devices to collaboratively train a joint machine learning model locally without sharing the individual training data. Most existing incentive mechanism schemes for FL assume that the task publisher is completely rational and capable of making decisions based on expected utility theory (EUT). However, in reality, the task publisher is characterized as bounded rationality under risks and uncertainties, whose risk-awareness makes EUT inapplicable when making decisions. To tackle the above challenge, a novel incentive mechanism for IIoT-FL based on contract theory and prospect theory is proposed in this paper. We leverage prospect theory to model the task publisher’s risk-awareness behavior. To guarantee high model accuracy while avoiding serious time delay, we take the global model quality and time satisfactory into account when designing the optimal contract. Simulation results demonstrate that our incentive mechanism is effective under asymmetric information and risk.