Blockchain-based Federated Learning Framework Applied in Face Recognition

The significant improvement of face recognition technology has mainly resulted from the rapid enhancement of Deep Neural Network performance and the use of large face datasets. As the use of face datasets concerns data privacy, it is difficult and unwilling for organizations and individuals to share their data for model training. Although existing federated learning methods can train data locally, the problems of the single point of failure and insufficient user incentives still exist. In response to the above issues, we propose an innovative BFLFace framework. BFLFace is a framework that uses blockchain instead the central server of federated learning to implement the decentralized training process of the face recognition model. At the same time, the combination of committee nodes that can evaluate the quality of the updated model and incentive mechanism based on blockchain that is developed to encourage more reliable users to participate in the training process can improve model performance. On the premise of ensuring the accuracy of the face recognition model, BFLFace can significantly solve the problem of data silos and further enhance the data security of users.