Topology Optimization Techniques for Decentralized Federated Learning: Challenges and Solutions

Decentralized Federated Learning (DFL) has emerged as a scalable and privacy-preserving framework for collaborative machine learning, enabling edge devices to train shared models without relying on a central server. However, the performance and convergence of DFL heavily depend on the underlying communication topology, which governs how clients exchange model updates. This study provides a comprehensive overview of topology optimization techniques in DFL, highlighting recent advances, challenges, and strategies aimed at enhancing communication efficiency, resilience, and learning performance. This study categorizes existing methods into heuristic, spectral, data-driven, and multi-objective approaches, analyzing their effectiveness in improving training performance under real-world constraints. Furthermore, the study identifies key research directions, including dynamic and adaptive topologies, learning-based approaches using graph neural networks, and privacy-preserving topology adaptation. This work serves as a foundational reference for researchers and practitioners seeking to advance scalable and intelligent DFL systems.