Felicitas: Federated Learning in Distributed Cross Device Collaborative Frameworks

Felicitas is a distributed cross-device Federated Learning (FL) framework to solve the industrial difficulties of FL in large-scale device deployment scenarios. In Felicitas, FL-Clients are deployed on mobile or embedded devices, while FL-Server is deployed on the cloud platform. We also summarize the challenges of FL deployment in industrial cross-device scenarios (massively parallel, stateless clients, non-use of client identifiers, highly unreliable, unsteady and complex deployment), and provide reliable solutions. We provide the source code and documents at https://www.mindspore.cn/. In addition, the Felicitas has been deployed on mobile phones in real world. At the end of the paper, we demonstrate the validity of the framework through experiments.