Communication-Efficient Device Scheduling for Federated Learning Using Stochastic Optimization

TL;DR

This paper introduces a stochastic optimization-based client selection algorithm for federated learning that reduces communication time in wireless environments without needing channel statistics, validated on FEMNIST and CIFAR-10 datasets.

Contribution

It provides a novel convergence analysis for non-convex federated learning and develops a channel-aware client selection algorithm that minimizes communication time.

Findings

Communication time significantly decreased with the proposed algorithm.

Algorithm does not require channel statistics, only instantaneous channel information.

Validated improvements on FEMNIST and CIFAR-10 datasets.

Abstract

Federated learning (FL) is a useful tool in distributed machine learning that utilizes users' local datasets in a privacy-preserving manner. When deploying FL in a constrained wireless environment; however, training models in a time-efficient manner can be a challenging task due to intermittent connectivity of devices, heterogeneous connection quality, and non-i.i.d. data. In this paper, we provide a novel convergence analysis of non-convex loss functions using FL on both i.i.d. and non-i.i.d. datasets with arbitrary device selection probabilities for each round. Then, using the derived convergence bound, we use stochastic optimization to develop a new client selection and power allocation algorithm that minimizes a function of the convergence bound and the average communication time under a transmit power constraint. We find an analytical solution to the minimization problem. One key feature of the algorithm is that knowledge of the channel statistics is not required and only the instantaneous channel state information needs to be known. Using the FEMNIST and CIFAR-10 datasets, we show through simulations that the communication time can be significantly decreased using our algorithm, compared to uniformly random participation.