A Comprehensive Comparison of Federated Learning Frameworks

Federated Learning is a Machine Learning paradigm that enables institutions to train models collaboratively on decentralized data without exchanging it. Training is performed locally on each participant’s device, and only model updates are shared and aggregated, rather than raw data. The selection of a suitable framework is a critical step in designing a Federated Learning System, as this choice significantly influences the capabilities and limitations of the system. While previous research has primarily focused on comparing Federated Learning functionalities, other important criteria, such as usability, technical performance, and legal aspects have often been addressed only partially or overlooked entirely. To address this gap and support informed framework selection, we conducted a comprehensive comparison of Federated Learning frameworks. Frameworks were identified through a structured literature review and an analysis of GitHub repositories. The comparison covers five relevant comparison criteria: (1) Federated Learning functionalities, (2) user-friendliness, (3) technical aspects, (4) legal aspects, and (5) performance evaluation.