Federated Learning Architectures: Centralized vs Decentralized Trade-offs

A new research paper experimentally compares centralized, decentralized, and semi-decentralized federated learning architectures using the Fedstellar simulator, MNIST dataset, and MLP classifier. The study addresses a gap in literature by analyzing trade-offs between performance indicators across these three FL types. Federated Learning enables collaborative model training on distributed edge devices while preserving data privacy, crucial for the growing number of IoT devices generating massive data. Centralized storage faces challenges like limited communication, privacy concerns, and regulations. The paper provides experimental comparisons to understand strengths, limitations, and performance trade-offs, helping practitioners choose the right architecture based on application needs.