Jupiter: A Modern Federated Learning Platform for Regional Medical Care

Download PDF
Open Webpage

Ju Xing, Zexun Jiang, Hao Yin

Published: 2020, Last Modified: 26 May 2026JCC 2020EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: In this paper we propose Jupiter, an easy-to-use, secure and high-performance federated learning platform for regional medical care. Jupiter provides innovative programming abstractions to make data tunning more efficient as well as high-performance infrastructures to accelerate secure aggregations of parameters. Jupiter employs a stateful design with a bunch of optimizations and leverages popular techniques like SDN, DPDK and Intel SGX. The experiments show that with a low memory footprint, the throughput of single aggregator can reach 300MB/sec(with slice size fixed to 64KB), and the aggregation primitive we built can process 11k aggregations per second.