TinyML-CoDE: A Co-Design and Evaluation Framework for Systematic TinyML Development

Abstract

Dhananjaya Lahiru Bandara, Mohamed Mirshad Munawwer and Charith Lakpriya Jayathilake

The current paper presents the key issues when developing Tiny Machine Learning in resource-constrained Internet of Things equipment. Based on the methodical research on 35 recent papers, we uncover five core research gaps that include standardized benchmarking, systematic co-design, lifecycle management, security integration, and toolchain interoperability. To implement these gaps, we suggest the TinyML Co-Design and Evaluation (TinyML-CoDE) framework to be based on the integrated approach to the methodology that considers constraint awareness, hardware- model recommendation, automated optimization, unified benchmarking, and lifecycle management. The benefits that are projected are 30 percent less development time and 25 percent performance efficiency. The framework offers a methodical way of eliminating the existing fragmentation and facilitating scalable deployment of TinyML in the industrial and healthcare IoT applications.

PDF