Hierarchical Federated Learning With Quantization: Convergence Analysis and System Design

Federated learning (FL) is a powerful distributed machine framework where server aggregates models trained by different clients without accessing their private data. Hierarchical FL, with client-edge-cloud aggregation hierarchy, can effectively leverage both the cloud server’s access to many clients’ data and edge servers’ closeness achieve high communication efficiency. Neural network quantization further reduce overhead during model uploading. To fully exploit advantages of hierarchical an accurate convergence analysis respect key system parameters needed. Unfortunately, existing loose does not consider quantization. In this paper, we derive tighter bound for FL The result leads practical guidelines important design problems such as client-edge edge-client association strategies. Based on obtained analytical results, optimize two intervals show that interval should slowly decay while edge-cloud needs adapt ratio propagation delay. Simulation results shall verify demonstrate effectiveness proposed strategy.