Byzantine-Robust Aggregation for Securing Decentralized Federated Learning

Researchers have proposed novel algorithms to enhance the security of Decentralized Federated Learning (DFL), a distributed machine learning approach that trains AI models locally on devices, eliminating the need for a central server.

Federated Learning (FL) addresses privacy concerns by training AI models locally on devices, according to a study published on arXiv^[1]. Decentralized Federated Learning (DFL) further enhances scalability and robustness by eliminating the central server. However, DFL faces challenges in optimizing security, as most Byzantine-robust algorithms are designed for centralized scenarios^[1]. To address this, researchers have proposed a novel Byzantine-robust aggregation algorithm, WFAgg, which employs multiple filters to identify and mitigate Byzantine attacks. Experimental results demonstrate the effectiveness of WFAgg in maintaining model accuracy and convergence in the presence of various Byzantine attack scenarios. Another framework, CO-DEFEND, enables multiple entities to collaboratively train a classification machine learning model for DoH threat detection while preserving data privacy and enhancing scalability^[2]. CO-DEFEND adapts four classical machine learning algorithms for federated scenarios and uses decision trees and random forests as model selection mechanisms, allowing each participant to retain interpretable and locally optimal decision structures.