Research On Worker Nodes Incentive Mechanism In Federated Learning System

In federated learning system,worker nodes need to consume a lot of computing and communication costs to participate in training tasks.Therefore,task publishers need to design incentive mechanism so that the worker nodes can obtain appropriate benefits.At present,many studies have used different technologies to evaluate the quality of workers’ nodes,so as to design a reasonable federal learning incentive mechanism.However,the existing studies are only for the single task publisher scenario,and there is a problem that the target incentive group is too concentrated.Therefore,this paper intends to realize a reliable federated learning incentive mechanism by studying the determination strategy of worker training compensation in multi task competition and the worker node task allocation strategy of node centralization.Firstly,aiming at the determination strategy of workers’ training compensation in multi task competition,a multi task incentive mechanism based on contract theory is proposed.Through studying the income function of task publisher and worker nodes in federated learning,the optimal contract design in multi task publisher scenario is solved.By comparing the contract design in the single task scenario,it is verified that the proposed optimal contract design scheme can meet the individual rationality constraints and incentive compatibility constraints at the same time in multi task competition scenario,so as to maximize the benefits of worker nodes and task publishers.Secondly,aiming at the task allocation strategy of worker nodes with centralized nodes,a worker node incentive mechanism based on reputation weight is proposed.The reputation weight is used to enlarge the quality discrimination between different worker nodes,and combined with the task selection algorithm to solve the reasonable task allocation strategy of worker nodes.Through comparing the quality distribution of workers’ nodes in different time periods in the system,it is verified that the proposed incentive mechanism can effectively encourage workers with different quality to stay in the federal learning system and prevent centralized high-quality nodes from colluding to damage the interests of task publishers.