Application of AHP for Weighting Clients in Federated Learning

Abstract

Federated Learning is a branch of Machine Learning. The main idea behind it, unlike traditional Machine Learning, is that it does not require data from the clients to create a global model, so clients keep their data private. Instead, clients train their model on their own devices and send their local model to the server, where the global model is aggregated and sent back to clients. In this research work, the Federated Averaging algorithm is modified so that clients get their weights by the Analytical Hierarchal Process. Results showed that applying AHP for weighting performed better than giving clients weights solely based on their dataset size, which the Federated Averaging algorithm does.