Optimizing Cluster Head Placement in Federated Clustering: A Genetic Algorithm Approach

Federated learning (FL) is a widely used distributed deep learning framework that protects the privacy of each client by exchanging model parameters instead of raw data. Federated clustering (FCL) is an area of research within FL focuses on grouping similar data that is globally distributed while maintaining the data locally. However, the deployment of federated clustering encounters increased communication costs due to the repetitive transmission of a large volume of model parameters between cluster members and cluster head server posing as global aggregator throughout the training cycle. This inefficiency poses new challenges including reduced updated model distribution speed due to high communication latency resulting in inflated communication cost due to increased resource consumption. The cluster head server placement plays a vital role in minimizing the network cost in federated clustering environment due to data transmission parameters such as network throughput,total bandwidth and average delay. The network cost minimization associated with head server placement problem in federated clustering is investigated. In this paper we propose genetic algorithm based cluster head placement technique (CHP) and minimize the average delay and total bandwidth required during federated learning rounds to achieve minimum network cost.