A Contract Theory Based Incentive Mechanism for Federated Learning

Federated learning (FL) serves as a data privacy-preserved machine paradigm, and realizes the collaborative model trained by distributed clients. To accomplish an FL task, task publisher needs to pay financial incentives server offloads contributing However, it is challenging design proper for clients because privately This paper proposes contract theory-based training toward minimizing incentive budget subject being individually rational compatible in each round. We two-dimensional formally defining two private types of clients, namely quality computation effort. effectively aggregate models, contract-based aggregator proposed. analyze feasible optimal solutions proposed model. The experimental results demonstrate that framework can improve generation accuracy tasks. Moreover, generalization tasks be improved mechanism, aggregation applied.