Multi-Objective Optimization for Bandwidth-Limited Federated Learning in Wireless Edge Systems

This paper studies a bandwidth-limited federated learning (FL) system where the access point is central server for aggregation and energy-constrained user equipemnts (UEs) with limited computation capabilities (e.g., Internet of Things devices) perform local training. Limited by bandwidth in wireless edge systems, only part UEs can participate each FL training round. Selecting different could affect performance, selected need to allocate their computing resource effectively. In simultaneously accelerating reducing computing-communication energy consumption are importance. To this end, we formulate multi-objective optimization problem (MOP). MOP, model convergence difficult calculate accurately. Meanwhile, MOP combinatorial problem, high-dimension mix-integer variables, which proved be NP-hard. address these challenges, evolutionary algorithm (MOEA-FL) proposed obtain Pareto optimal solution set. MOEA-FL, an age-of-update-loss method first transform original global loss function into reference function. Then, MOEA-FL divides N single objective subproblems Tchebycheff approach optimizes evolving population. Extensive experiments have been carried out on MNIST dataset medical case called TissueMNIST both i.i.d non-i.i.d data setting. Experimental results demonstrate that performs better than other algorithms verify robustness scalability MOEA-FL.