FLRA: A Reference Architecture for Federated Learning Systems

Federated learning is an emerging machine paradigm that enables multiple devices to train models locally and formulate a global model, without sharing the clients’ local data. A federated system can be viewed as large-scale distributed system, involving different components stakeholders with diverse requirements constraints. Hence, developing requires both software design thinking knowledge. Although much effort has been put into from perspectives, our previous systematic literature review on area shows there distinct lack of considerations for architecture learning. In this paper, we propose FLRA, reference systems, which provides template learning-based solutions. The proposed FLRA based extensive existing patterns systems found in industrial implementation. consists pool architectural could address frequently recurring problems architectures. serve guideline assist architects developers practical solutions their problems, further customised.