On the Selection of a Machine Learning model in TinyML Devices – Preliminary Study

Abstract

The expected development of TinyML-related technologies will necessitate the development of methods for efficient use of energy resources. In this article, we present preliminary study of machine learning (ML) model selection in TinyML devices in order to reach a tradeoff between accuracy and energy consumption. We study various use cases with different ML models. Our research shows that the presented method can improve the TinyML system in terms of operation time at the cost of slightly lower accuracy.