Blockchain Empowered Federated Learning for Data Sharing Incentive Mechanism

Abstract

In the machine learning, data sharing between different participants can increase the amount of data, improve the quality of the dataset, and thereby improve the quality of the model. Under the condition of data supervision, federated learning, as a distributed machine learning, aims to protect data while training models through collaboration among all parties to achieve data sharing and improve model quality. However, there are still some issues. For instance, the lack of trust between the participants makes it impossible to establish a secure and reliable sharing mechanism. In addition, how to fairly share the benefits generated by the model, identify honest participants and punish malicious participants is still a challenge. In this paper, we propose a new federated learning scheme based on blockchain architecture for federated learning data sharing. Moreover, an incentive mechanism based on reputation points and Shaply values is proposed to improve the sustainability of the federated learning system, which provides a credible participation mechanism for data sharing based on federated learning and fair incentives. The experimental results and analysis show that the loss of federated learning is more smooth than that of centralized machine learning.