Enhancing Model Robustness in Federated Learning: A Systematic Literature Review of Byzantine-Resilient Aggregation Methods

The demand for privacy-preserving machine learning has led to the rise of Federated Learning (FL), where multiple clients collaboratively train a model without sharing raw data. Despite its privacy benefits, FL is vulnerable to Byzantine failures, where malicious or faulty participants inject corrupted updates, threatening model integrity. To address this, a range of Byzantine-resilient aggregation techniques have been proposed, including statistical filters (e.g., Trimmed Mean, Krum), trust-based weighting, cryptographic protocols, and hybrid strategies. This paper presents a systematic literature review (SLR) of these defenses, evaluating their robustness, scalability, and suitability for real-world applications. Challenges such as non-IID data, adaptive attacks, and trade-offs between security and efficiency are critically examined. In addition, we explore emerging trends such as domain-specific defenses, energy-aware FL, quantum-resilient methods, and federated zero-knowledge proofs. A novel classification of hybrid approaches and a standardized benchmarking framework are proposed to guide future research. This review aims to support the development of resilient, efficient and scalable decentralized learning systems in adversarial environments.