FedDdrl: Federated Double Deep Reinforcement Learning for Heterogeneous IoT with Adaptive Early Client Termination...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Distributed Systems

Background:

• Federated learning (FL) enables collaborative model training without data sharing, crucial for privacy and bandwidth constraints.
• Heterogeneous Internet of Things (IoT) environments present challenges like non-independent and identically distributed (non-IID) data and varied device capabilities.
• Optimizing FL involves balancing global model accuracy, training latency, and communication costs.

Purpose of the Study:

• To propose a joint early client termination and local epoch adjustment strategy for federated learning.
• To address the challenges posed by heterogeneous IoT environments in FL.
• To achieve an optimal tradeoff between model accuracy, training latency, and communication cost.

Main Methods:

• Mitigation of non-IID data influence using the balanced-MixUp technique.
• Formulation of a weighted sum optimization problem.
• Development of a federated learning double deep reinforcement learning (FedDdrl) framework for dual action output (client termination and epoch adjustment).

Main Results:

• The FedDdrl framework successfully balances conflicting objectives in FL.
• FedDdrl demonstrates superior performance over existing FL schemes in terms of overall tradeoff.
• Achieved approximately 4% higher model accuracy with 30% less latency and communication costs.

Conclusions:

• The proposed FedDdrl framework effectively enhances federated learning performance in heterogeneous IoT settings.
• Joint early client termination and local epoch adjustment are viable strategies for optimizing FL.
• The approach offers significant improvements in accuracy, latency, and communication efficiency.