Incentive mechanism for collaborative distributed learning in Artificial Intelligence of Things

Abstract

Artificial Intelligence of Things (AIoT) is an emerging trend that integrates artificial intelligence into the Internet of Things, which enables intelligent IoT operations and smart industrial applications. AIoT can generate a large amount of data from the network edge. Due to the concerns of bandwidth and privacy, it is often impractical to move the collected data to the cloud. To address this issue, a collaborative distributed learning has been proposed to let the clients collaboratively train a machine learning model together with their local data in a distributed manner. In this paper, we study the free market incentive mechanism for collaborative distributed learning, where multiple parameter servers (PSs) compete with each other to motivate clients to contribute model training. Specifically, the Stackelberg game with multiple leaders and multiple followers has been designed to analyze the incentive mechanism. Different experiments have been performed to illustrate the efficiency of the proposed approach. In particular, compared with the state-of-the-art under the same budget constraint, the final average utility of the PSs can be increased by at least 94.3%.