Enhancing On-Device learning in IoT Systems through Meta-Learning Techniques: A Comprehensive review

Abstract

The incorporation of meta-learning approaches to on-device learning for IoT systems has emerged as one of the effective ways of developing intelligent and never-stopping devices capable of learning and adapting on their own. This paper aims to examine existing literature to highlight the progress, prospects, and potential complications prevalent in this dynamic field. The paper reviews on specialized hardware architecture, meta-learning algorithms, and system modularity that support on-device learning in constrained IoT systems. This study investigates several existing methods to enhance ondevice learning, like as Federated Learning (FL), Transfer Learning (TL), and Continual Learning (CL) in relation to IoT systems by using meta-learning. We also cover the predictive modeling perspectives, performance assessment, and emerging issues such as privacy, security, and professional ethics. Thus, this review synthesizes latest research works and current literature to identify gaps of existing knowledge to enhance on-device learning in IoT systems through metalearning techniques. This enables researchers and practitioners to get insights with a comprehensive understanding of the state-of-the-art, future prospects and potential developments of on-device meta-learning in IoT systems, fostering further advancements in this rapidly evolving area of study.