Research On Multidimensional Contract-based Incentive Mechanism For Federated Learning In Competitive Environment

Federated learning is an emerging technique for bridging data silos and privacy protection,and it has garnered considerable attention from both academia and industry.The key to implement this technology is the incentive mechanism which motivates data owners to participate in a federated learning.In a competitive environment where there is a competitive relationship between a task publisher and some data owners,conventional incentive mechanisms are difficult to effectively motivate data owners to participate in federated learning.Hence,to effectively motivate data owners in a competitive environment to participate in federated learning,a multi-dimensional contract-based incentive mechanism in a competitive environment is proposed.The main contents of this thesis are presented as follows:First,problem analysis and the solution approach are provided.Firstly,the reasons for the implementation of federated learning in a competitive environment are analyzed from objective and subjective factors.Then,the characteristics of federated learning in a competitive environment are summarized from the perspectives of federated learning algorithms and participants.Further,to design the incentive mechanism for federated learning in a competitive environment,an overall solution framework is given.Finally,specific solutions to the design problem of the multi-dimensional contract-based incentive mechanism for federated learning based on complete and incomplete information are respectively provided.Second,the multi-dimensional contract-based incentive mechanism for competitive federated learning based on complete information is designed.Firstly,the competition intensity is introduced to quantify the competition relationship between data owners and the task publisher in a competitive environment,and the profit functions of the task publisher and data owners in a competitive environment are respectively given.Then,a novel incentive form named MM(Monetary-the Model Usage)combined incentive,which combines monetary with the usage of the federated learning model,is proposed.Thirdly,based on the MM combined incentive,the design optimization problem of the multi-dimensional contract-based incentive mechanism for federated learning in a competitive environment based on complete information is constructed.Further,to solve the constructed optimization problem,the relation between the optimal solutions of the optimization problem is analyzed based on complete information.Finally,the feasibility and superiority of the multi-dimensional contract-based incentive mechanism based on complete information are verified by a numerical example.Third,the multi-dimensional contract-based incentive mechanism for competitive federated learning based on incomplete information is designed.Firstly,based on incomplete information,the profit functions of the task publisher and data owners in a competitive environment are respectively given.Then,the design optimization problem of the multi-dimensional contract-based incentive mechanism for competitive federated learning based on incomplete information is constructed.Thirdly,to solve the constructed optimization problem,feasibility and optimality analysis of contracts in the incentive mechanism is analyzed.Finally,the feasibility and superiority of the multi-dimensional contract-based incentive mechanism based on incomplete information are verified by a numerical example.This thesis further enriches the theoretical research of the incentive mechanism for federated learning and provides a mechanism guarantee for effectively implementing federated learning in a competitive environment.