Distributed Joint Resource Optimization for Federated Learning Task Distribution

Abstract

The effective utilization of computation and communication resources is one of the primary challenges for Artificial Intelligence (AI)-based applications. The complexity of the edge computation environment and the diversity of communication interference in the wireless environment make it particularly difficult to quickly and accurately complete multiple federated learning tasks. This paper examines energy cost, communication capacity, computation capacity, and other factors in a network-slicing environment. The virtual queue vector and the associated Lyapunov drift are presented to analyze queue backlog performance and optimize each slice’s Key Performance Indicator (KPI) performance, including energy and other Quality of Experience (QoE) factors. In response, we design a Federated Learning (FL) Service-Oriented Allocation Policy Calculation Algorithm to address the allocation policy calculation problem in a time-varying environment. Finally, our simulation experiments demonstrate that our proposed algorithm outperforms other benchmarks.