APONS: Accelerating Federated Learning Architecture in 6G Based on Parameter Optimization and Neural Architecture Search
Weisen Pan, Jian Li, Liangyu Gao, S Bao, Quan Zhao, Qixing Wang, Chunfeng Cui · 2023
The Federated Edge Learning technique can successfully assist the edge deployment of 6G networks. Using a lot of user data, it can train machine learning models by communicating with many edge clients. However, in 6G-enabled mobile edge computing networks, heterogeneity and resource limitations among distributed edge clients can lower the effectiveness of Federated Edge Learning training. This study suggests a novel Federated Learning framework to expedite the training process. This paper proposes a new model based on the relationship between training loss, resource consumption, and heterogeneity. Then, to reduce latency effects brought on by client heterogeneity and resource limitations, we suggest using a search technique called APONS to generate local models of edge clients and optimal imprecision of band allocation. As a result, modifying the percentage of frequency bands and the local model of the edge client's inaccuracy can significantly increase training efficiency. The simulation outcomes demonstrate our algorithm's benefits in increasing training effectiveness.