Federated learning using game strategies: State-of-the-art and future trends

Federated learning (FL) is a new and promising paradigm that allows devices to learn without sharing data with the centralized server. It is often built on decentralized data where edge nodes use the internet of everything to mitigate the malicious attacks. The server gives incentive to all the participants according to their individual contributions. For profit maximization, each participating node balances between training rewards and costs. Game theory (GT) is a mathematical optimization technique that can be used to solve problems in wireless communication including security, resource allocation, power management, node rewards and punishments, and balancing numerous trade-offs. The complicated interactions between the server and the edge devices are interpreted using GT to maximize their utility. In this review article, we present an overview of the latest research on GT-based FL models for profit maximization, authentication, privacy management, trust management, and threat detection. This study also investigates the bibliometric analysis covering the period from 2019 to 2022 with an emphasis on various mechanisms of FL for GT applications. This article seeks to fill the gap by exploring the significant works highlighting the authors, citations, algorithms used, findings, and applications in this field. Based on the findings, we conclude this article with several key challenges and future approaches for researchers to implement an efficient GT-based FL model.