Federated Learning: The Effect of Device Clustering for Multi-hop Networks

Federated Learning (FL) is a cooperative machine learning (ML) training method that involves several participants in the training phase. The learning phase of ML algorithms is done in a distributed manner where devices train a specific model based on their local data sets and send the updated model back to a centralized entity. Several benefits are gained by using FL compared to conventional centralized learning techniques, such as increased security and data privacy since participant devices only send back the updated model and not their data sets. Recently, this has become a hot research topic due to its advantages in terms of user privacy. Most of the works in this field consider only star networks where mobile devices communicate directly to the base station. Few mentioned multi-hop scenarios. In this paper, we present a performance evaluation of FL in multi-hop networks and study the effect of clustering devices on FL accuracy.