Federated learning on non-iid data using continual learning method

Abstract

Federated learning is a method that can protect data privacy of local devices and at the same time reduce communication costs, and is attracting great interest from many researchers due to these advantages. It is confirmed that the performance of this federated learning is lower than in the case of the iid situation (the data are all distributed equally) in the non-iid situation where the local devices are non-independent non-identical data.In this paper, we present a method of applying the continuous learning method to federated learning to solve the poor performance of federated learning in non-iid data situations. We introduce a method that can achieve better performance than the vanilla federated learning method in non-iid data situations by introducing a continuous learning method to the learning method of local device while maintaining the data setting and learning method in the vanilla federated learning.