Convergence Analysis and System Design for Federated Learning over Wireless Networks

Abstract

Federated learning (FL) has recently emerged as an important and promising learning scheme in IoT, enabling devices to jointly learn a model without sharing their raw data sets. However, as the training data in FL is not collected and stored centrally, FL training requires frequent model exchange, which is largely affected by the wireless communication network. Therein, limited bandwidth and random package loss restrict interactions in training. Meanwhile, the insufficient message synchronization among distributed clients could also affect FL convergence. In this paper, we analyze the convergence rate of FL training considering the joint impact of communication network and training settings. Further by considering the training costs in terms of time and power, the optimal scheduling problems for communication networks are formulated. The developed theoretical results can be used to assist the system parameter selections and explain the principle of how the wireless communication system could influence the distributed training process and network scheduling.

Training Time Minimization for Federated Edge Learning with Optimized Gradient Quantization and Bandwidth Allocation

“…3) Minimization of total training time: The work in [22] minimized the total training time by optimizing communication and computation resource allocation; however, no compression was considered in [22]. The authors in [7] minimized the training time for a fixed communication rounds by solving a joint learning, wireless resource allocation, and device selection problem.…”

“…Besides, the edge devices in the edge network are usually heterogeneous such that some of edge devices with lower computation power become laggards in the synchronous model/gradient aggregation due to their longer computation time, which increases the per-round latency. It is necessary to consider the optimization of wireless resources over different edge devices to reduce the communication time of the lagging edge devices, and compensate the longer computation time [20]- [22]. With the goal of minimizing the total training time, we study the following question: How to balance the trade-off between the number of communication rounds and per-round latency via joint quantization level and bandwidth allocation optimization in the presence of device heterogeneity.…”