Basic Consideration of Federated Rough Set C-Means Clustering and Its Application to Collaborative Filtering

Abstract

In recent years, as the utilization of big data is increasingly demanded, the importance of clustering, which is a technique for automatically classifying and summarizing data without supervision, is growing. Rough clustering, which incorporates the perspective of rough set theory, is a clustering method that handles the uncertainty of each object’s membership to each cluster. Furthermore, federated learning is a method for collaboratively training models in a distributed environment without aggregating data held by multiple clients onto a server. Federated learning is expected to provide privacy protection, reduce communication costs, and enable effective utilization of computational resources, among other benefits. As an extension of the representative rough clustering method, rough C-means (RCM), to federated learning, federated RCM (F-RCM) has been proposed. In this study, we extend rough set C-means (RSCM), a rough clustering method that considers granularity, to federated learning and propose federated RSCM (F-RSCM). Additionally, for benchmarking, we apply each method to a collaborative filtering task and compare their recommendation performance.