Improving Communication Efficiency in IoT-based Federated Machine Learning Systems Using Software Defined Networks

This article focuses on the medical federated machine learning system where clients are portable IoT devices. Currently, multiple limitations complicate the implementation and use of similar systems. In this article, we will address the problem of communication overhead [1] considering the utilization of IoT devices, which typically rely on wireless communication technologies for data transfer, and the size of this data transfer, which usually is larger than double the size of the neural network model being used (might reach hundreds of megabytes). To mitigate the described challenge, we propose a solution based on Software Define Networks, which uses a set of routers and a server. The provided solution will lower the risk of unexpected connection losses and other network errors producing communication overhead in the described system, thus boosting the overall system performance during communication and providing other advantages regarding the quality of the provided communication link.