Device Scheduling and Update Aggregation Policies for Asynchronous Federated Learning
Title: |
Device Scheduling and Update Aggregation Policies for Asynchronous Federated Learning |
DNr: |
SNIC 2021/22-296 |
Project Type: |
SNIC Small Compute |
Principal Investigator: |
Chung-Hsuan Hu <chung-hsuan.hu@liu.se> |
Affiliation: |
Linköpings universitet |
Duration: |
2021-04-16 – 2022-05-01 |
Classification: |
20203 |
Keywords: |
|
Abstract
Federated learning (FL) is a decentralized collaborative learning framework with multiple agents participating in a common training process iteratively over locally distributed data. There are common issues in FL systems such as stringent communication costs in wireless communication. In this work, we propose several scheduling and aggregation policies for FL system to resolve the issue and guarantee convergence of the system.