Towards Federated Learning at Scale: System Design

TL;DR

The paper tackles scalable Federated Learning on consumer devices, addressing privacy and coordination challenges when training on decentralized mobile data. It presents a production-efficient system built on TensorFlow with synchronous rounds, a robust Actor Model server, device attestation, and secure aggregation. The work reports on a deployed fleet reaching tens of millions of devices and demonstrates practical applications such as on-device keyboard ranking and next-word prediction with live metrics. It also discusses analytics, tooling, and open research directions including bias, parallelism beyond hundreds of devices, and broader Federated Computation opportunities.

Abstract

Federated Learning is a distributed machine learning approach which enables model training on a large corpus of decentralized data. We have built a scalable production system for Federated Learning in the domain of mobile devices, based on TensorFlow. In this paper, we describe the resulting high-level design, sketch some of the challenges and their solutions, and touch upon the open problems and future directions.

Figures (9)

• Figure 1: Federated Learning Protocol
• Figure 2: Device Architecture
• Figure 3: Actors in the FL Server Architecture
• Figure 4: Model Engineer Workflow
• Figure 5: Round Completion Rate