Shortest path core-selection incentive for federated learning platform with medical applications

As the main technology to solve data islands and mine data value, federated learning has been widely researched and applied, and more and more federated learning platforms are emerging. The federated learning platform organizes users, devices and data to collaborate in a crowdsourcing manner and complete specific federated learning tasks. This paper designs the shortest path core-selection incentive mechanism by combining the VCG auction mechanism and the core concept of cooperative games. This mechanism solves the problems of overpayment, false-name attack, and deviation from the core of the VCG mechanism, and saves the expenditure of the federated learning task issuer. It adopts the VCG-nearest principle in the core selection, so that the federated learning task participants get rewards as close as possible to the outcome of VCG mechanism. This mechanism can guarantee four economic attributes: incentive compatibility, individual rationality, alliance rationality, and maximization of social efficiency. Medical experimental results illustrate the effectiveness of the mechanism.