wFedUFCM: A Federated Approach to Weighted Unconstrained Fuzzy C-Means Clustering
Longmei Li, Wei Lu · 2025
In recent years, there has been a growing need for federated clustering methodologies in real-world applications that prioritize privacy considerations. This study proposes a novel generalized federated embedding clustering method, termed Weighted Unconstrained Deep Fuzzy C-Means (wFedUFCM). The technique integrates an autoencoder (AE) and an UFCMN to learn both feature representations and cluster assignments simultaneously. The principal advantage of wFedUFCM lies in its use of a weighted UFCMN (wUFCMN) for clustering. Unlike traditional methods that involve a cumbersome optimization process with alternating updates, wUFCMN employs gradient descent to optimize the objective function for clustering loss, enabling end-to-end optimization through parameter updates via gradient descent and backpropagation. Considering that data is often nonIndependent and Identically Distributed (non-IID), we address this challenge by utilizing simultaneous model and center averaging. Specifically, model averaging is achieved through a federated approach, while centers are reassigned using k-means clustering to update global centers effectively.