Research On Node Incentive Mechanism For Hierarchical Federated Learnin

Traditional federated learning has a long distance between the mobile terminal and the parameter server,resulting in low efficiency in model training.In the hierarchical federated learning(HFL)combined with mobile edge computing,edge servers are usually deployed as mediators on the base station between mobile devices and the cloud,and can perform edge aggregation of local models transmitted from nearby devices.In order to improve the quality of the global model and the participation of mobile devices and edge servers,designing effective incentive mechanisms is a key research.Firstly,energy constrained mobile devices will consume their own resources for participate in model training in HFL.In order to reduce the energy consumption of mobile devices,we propose the problem of minimizing the sum of energy consumption of mobile devices without exceeding the maximum tolerance time of HFL.Different training rounds of edge servers can select different mobile devices,and mobile devices can also train models under different edge servers concurrently.Therefore,we propose ODAM-DS algorithm based on an online double auction mechanism.Based on the optimal stopping theory,edge servers are supported to select the mobile device at the best time,so as to minimize the average energy consumption of mobile devices.Then,the theoretical analysis of the proposed ODAM-DS algorithm proves that it meets the characteristics of incentive compatibility,individual rationality and weak budget equilibrium constraints.Simulation results show that the energy consumption of ODAM-DS algorithm is 19.04% lower than that of the existing HFEL algorithm.Secondly,we investigate the aggregated model quality maximization problem in HFL,the decision problem of which is proved NP-complete.We develop the mechanism Max Q to maximize the sum of local model quality,which consists of two stages.In the first stage,an algorithm based on matching game theory is proposed to associate mobile devices with edge servers,which is proved able to achieve the stability and 1/2-approximation ratio.In the second stage,we design an incentive mechanism based on contract theory to maximize the quality of models submitted by mobile devices to edge servers.Through thorough experiments,we analyze the performance of Max Q and compare it with the existing mechanisms FAIR and EHFL,under different deep learning models Res Net18,Res Net50 and Alex Net,individually.It is found that the model quality can be improved by 8.20% and7.81%,10.47% and 11.87%,10.98% and 11.97% under different models,respectively.Thirdly,in HFL,the social welfare maximization problem is proposed by introducing model quality in the utility function in order to improve the global model quality.Based on Pareto optimality it is transformed into a multi-objective optimization problem that maximizes the utility function for the cloud server,edge servers and devices.In order to better motivate the participation of the cloud server,edge servers and devices,the utility of all the cloud server,edge servers and devices is maximized by solving the Pareto fair solution.In the process of solving the Pareto fair solution,the algorithm PFS is designed to find the Pareto fair solution by using the properties of the weighting coefficients of the objective function.In terms of model aggregation,the model quality is used as the aggregation weight to improve the quality of the aggregated model.Through simulation experimental analysis,the PFS algorithm can effectively speed up the model convergence and improve the social welfare of the model by 7.58%,16.87% and 7.02%,15.17% on the CIFAR10 and MNIST datasets,respectively,compared with the existing POS and NES.