Queuing dynamics of asynchronous Federated Learning

TL;DR

This paper analyzes the queuing dynamics in asynchronous federated learning with heterogeneous node speeds, proposing a non-uniform sampling scheme that improves delay and complexity by leveraging the system's queuing structure, leading to better performance.

Contribution

It introduces a novel non-uniform sampling method for asynchronous federated learning that accounts for queuing dynamics, improving efficiency over existing algorithms.

Findings

Significant delay reduction in federated learning systems.

Improved convergence complexity compared to state-of-the-art methods.

Enhanced image classification performance in experiments.

Abstract

We study asynchronous federated learning mechanisms with nodes having potentially different computational speeds. In such an environment, each node is allowed to work on models with potential delays and contribute to updates to the central server at its own pace. Existing analyses of such algorithms typically depend on intractable quantities such as the maximum node delay and do not consider the underlying queuing dynamics of the system. In this paper, we propose a non-uniform sampling scheme for the central server that allows for lower delays with better complexity, taking into account the closed Jackson network structure of the associated computational graph. Our experiments clearly show a significant improvement of our method over current state-of-the-art asynchronous algorithms on an image classification problem.