Client Selection Approach in Support of Clustered Federated Learning over Wireless Edge Networks

Clustered Federated Multitask Learning (CFL) was introduced as an efficient\nscheme to obtain reliable specialized models when data is imbalanced and\ndistributed in a non-i.i.d. (non-independent and identically distributed)\nfashion amongst clients. While a similarity measure metric, like the cosine\nsimilarity, can be used to endow groups of the client with a specialized model,\nthis process can be arduous as the server should involve all clients in each of\nthe federated learning rounds. Therefore, it is imperative that a subset of\nclients is selected periodically due to the limited bandwidth and latency\nconstraints at the network edge. To this end, this paper proposes a new client\nselection algorithm that aims to accelerate the convergence rate for obtaining\nspecialized machine learning models that achieve high test accuracies for all\nclient groups. Specifically, we introduce a client selection approach that\nleverages the devices' heterogeneity to schedule the clients based on their\nround latency and exploits the bandwidth reuse for clients that consume more\ntime to update the model. Then, the server performs model averaging and\nclusters the clients based on predefined thresholds. When a specific cluster\nreaches a stationary point, the proposed algorithm uses a greedy scheduling\nalgorithm for that group by selecting the clients with less latency to update\nthe model. Extensive experiments show that the proposed approach lowers the\ntraining time and accelerates the convergence rate by up to 50% while imbuing\neach client with a specialized model that is fit for its local data\ndistribution.\n