Blockchain enabled secure federated learning framework

Abstract

Federate machine learning (FML) is a novel concept that trains the model to leverage data from many users rather than store the data. Federated learning (FL) allows participants to be involved without disclosing sensitive data to train the model. The server will initialize the global model with all connected participants. After the initialization, the initial global model gets trained locally with the participant’s local data set. The level of security directly affects or impacts the overall performance of the FML. Also, many security frameworks in FML are designed to handle specific types of attacks in the training phase, communication phase, or aggregation phase. Integrating Blockchain into FML system would greatly help to enhance the security further. Therefore, this work propose a Convolution Neural Network (CNN) based novel Blockchain enabled secure federated learning method to leverage security benefits for image processing applications and benchmark the performance in terms of running time for key generation in authentication, global model generation in the server, the model accuracy and loss. The proposed scheme is suitable for generic image processing applications in Healthcare, Agriculture, Face detection etc.