.png){kind=icon}
.png){kind=icon}
Return
An Efficient Joint Source-Channel Coding Scheme for Wireless Hierarchical Federated Learning and its Information-Theoretic Analysis
.png){kind=icon}
PRE
AI
DOI：10.1109/TCE.2024.3432178.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Wireless hierarchical federated learning (WHFL) is an implementation of wireless federated learning (WFL) on a cloud-edge-client hierarchical architecture that accelerates model training and achieves more favorable trade-offs between communication and computation. However, the unreliable wireless channel causes aggregation distortion and transmission latency, which significantly influence the learning performance of WHFL. To solve these problems, this paper proposes a feedback oriented joint source-channel coding (FO-JSCC) scheme for the multiple-input multiple-output (MIMO) WHFL system, which exploits channel output feedback, and provides significant improvements in gradient distortion under fixed coding blocklength or in transmission latency under target gradient distortion. Besides this, we analyze the impact of gradient distortion on the convergence rate and learning efficiency of WHFL. Simulation results show that our proposed scheme reduces model gradient distortion and enhances learning performance, and the transmission latency of our proposed scheme is about 2-4 times lower than that of traditional coding schemes.
Keywords:
Servers
Wireless communication
Encoding
Distortion
Wireless sensor networks
Federated learning
Communication system security
Channel feedback
hierarchical federated learning
joint source-channel coding
wireless edge computing
Journal
IF:
10.9
Papers:
5.1K
Citations:
6.8K