Strategies for Reducing the Communication and Computation Costs in Cross-Silo Federated Learning: A Comprehensive Review

Federated learning is an innovative approach that allows collaboration across distributed clients while maintaining data privacy. Despite its numerous benefits, several issues persist in this domain. This comprehensive review examines the critical problems of communication and computation (C&C) costs in cross-silo federated learning environments, which significantly impact the scalability and practical adoption of the system. The research presents a novel multi-dimensional analysis methodology to evaluate cost reduction techniques that consider privacy, accuracy, scalability, and adaptability. The methodical investigation reveals that while existing approaches can substantially reduce communication overhead in controlled environments, ensuring model convergence and privacy guarantees remains challenging across diverse scenarios. Through detailed case studies spanning smart city deployments, healthcare, and finance sectors, the study demonstrates how various C&C optimization strategies perform differently in real-world applications. The review introduces a systematic taxonomy of cost-reduction techniques and proves that hybrid approaches combining multiple optimization methods can maintain model performance while optimizing resource utilization. The review concludes by presenting a unified roadmap for developing adaptive solutions that balance privacy, efficiency, and scalability requirements in cross-silo contexts and identifying crucial research gaps. This compilation of recent developments focuses on areas that require further investigation to enhance real-world deployments and provides practitioners with actionable guidelines for implementing effective federated learning systems.