Fuzzy-based Augmentation of Federated Averaging for Enhanced Decentralized Machine Learning

Abstract

Federated Averaging (FedAvg) is a leading decentralized machine learning approach, prioritizing data privacy. However, it faces challenges like non-identically distributed data, communication bottlenecks, and adversarial attacks. This abstract introduces a fuzzy-based FedAvg, leveraging fuzzy logic to manage uncertainty in decentralized environments. Fuzzy clustering adapts the model to varied data distributions, addressing non-IID challenges. Fuzzy membership functions enhance aggregation by introducing an adaptive weighting scheme, improving convergence and accuracy. The fuzzy approach incorporates privacy-preserving mechanisms, ensuring secure aggregation with homomorphic encryption and differential privacy. Simulations show improved convergence, resilience to non-IID data, and enhanced privacy compared to traditional FedAvg, contributing to more secure decentralized ML systems.