Heterogeneity-aware high-efficiency federated learning with hybrid synchronous-asynchronous splitting strategy

This paper proposes a resource-aware adaptive model customization mechanism for federated learning that achieves faster convergence, higher accuracy, and lower network traffic.

Authors: Zijian Li, Boyuan Li, Kunyu Zhang, Bingcai Wei, Hongbo Liu, Zihan Chen, Xinqiang Xie, Tony Q.S. Quek

Key Contributions:

• Adapt resource-aware adaptive model customization mechanism tailor suitable model
• Mix synchronous edge feature extraction with asynchronous global classifier updates
• Use baseline-prioritized weighted aggregation to balance model update directions
• Achieve faster convergence, higher accuracy, and lower network traffic

Links: [Paper]