IEEE Communications Letters · 2022 · 84 citations · 12 references
ComputingComputer ScienceComputer SystemsEdge ComputingMachine LearningMobile SecurityEdge Artificial IntelligenceEdge SecurityFederated LearningIntrusion DetectionMobile Edge ComputingTransfer LearningAccuracyClient SelectionCommunication EfficiencyEfficiencyFederated Transfer LearningMobile Edge Computing ArchitectureReinforcement Learning TransferTraining
In this letter, we propose an efficient federated transfer learning (FTL) framework with client selection for intrusion detection (ID) in mobile edge computing (MEC). Specifically, we leverage federated learning (FL) to preserve privacy by training model locally, and utilize transfer learning (TL) to improve training efficiency by knowledge transfer. For FL, unreliable and low-quality clients should not be selected to participate in the training. Therefore, we integrate FTL with a reinforcement learning (RL)-based client selection scheme to achieve the highest ID accuracy within a budget limit on the number of participating clients. Experimental results show that the FTL significantly improves ID accuracy and communication efficiency as compared with the FL. Furthermore, the FTL framework with RL-based client selection can achieve the highest accuracy within budget, which improves performance while saving cost.
12
Optimizing Federated Learning on Non-IID Data with Reinforcement Learning
Hao Wang, Zakhary Kaplan, Di Niu et al. · 2020 · 958 citations