Year: 2026
Type: article
Abstract: Federated learning enables multiple clients to collaboratively train a shared machine-learning model without directly poexchanging their raw data. In distributed implementations, Ring-All-Reduce is commonly used to aggregate model updates efficiently, but its standard aggregation operations can be vulnerable when one or more participants send corrupted, manipulated, or intentionally malicious updates. This paper presents a conceptual framework for combining Byzantine-resilient update validation validation with Ring-All-Reduce communication. The proposed design uses lightweight local checks, robust aggregation at designated validation points, consistency monitoring, and recovery procedures to reduce the influence of faulty participants while retaining the communication advantages of ring-based training. The paper describes the problem, reviews relevant concepts, outlines the architecture and algorithm, and discusses evaluation metrics and practical limitations. It does not claim experimental results; instead, it provides a reproducible evaluation plan for comparing the proposed approach with ordinary Ring-All-Reduce an aggregation baselines. The central objective is to improve the reliability of distributed federated training under adversarial or unreliable client behaviour without unnecessarily exposing private training data.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7224123942