Research Article
On the Performance of Federated Learning Network
@INPROCEEDINGS{10.1007/978-3-031-54531-3_3, author={Godwin Idoje and Tasos Dagiuklas and Muddesar Iqbal}, title={On the Performance of Federated Learning Network}, proceedings={Collaborative Computing: Networking, Applications and Worksharing. 19th EAI International Conference, CollaborateCom 2023, Corfu Island, Greece, October 4-6, 2023, Proceedings, Part III}, proceedings_a={COLLABORATECOM PART 3}, year={2024}, month={2}, keywords={Federated Learning Bandwidth delay product aggregate model Round Trip time Traffic speed}, doi={10.1007/978-3-031-54531-3_3} }- Godwin Idoje
Tasos Dagiuklas
Muddesar Iqbal
Year: 2024
On the Performance of Federated Learning Network
COLLABORATECOM PART 3
Springer
DOI: 10.1007/978-3-031-54531-3_3
Abstract
Federated Learning is a decentralised network platform where the edge nodes train their local models and send their updated weights to the server. The server combines all the various local weights received and sends the aggregated model back to the edge nodes for further training, and this process continues until convergence is achieved. This study models the Federated Learning (FL) network. The Traffic speed (TS), Round trip time (RTT), and Bandwidth delay-product (BDP) parameters have been considered for modelling the Federated Learning network. Through experimentation, it can be inferred that the TS has a high impact and high correlation on the BDP within the network, and the RTT has a low impact on the BDP. The decentralised and classical machine learning models’ predictions have been compared. It has been observed that the decentralised machine learning model’s prediction outperforms the classical machine learning model’s prediction. The link experiences low latency because only the updated weights are transmitted within the link and not the raw data.
