Edge Machine Learning for AI-Enabled IoT Devices: A Review

Area of Science:

• Computer Science
• Electrical Engineering
• Artificial Intelligence

Background:

• The proliferation of connected devices in homes, cities, and industries necessitates efficient data processing.
• Limited resources on many Internet of Things (IoT) devices pose challenges for complex machine learning (ML) models.
• Network congestion is a significant bottleneck for large-scale IoT deployments.

Purpose of the Study:

• To review techniques enabling machine learning execution on low-performance IoT hardware.
• To define the state-of-the-art in edge machine learning for the IoT paradigm.
• To envision future development requirements for the Internet of Conscious Things.

Main Methods:

• Comprehensive literature review of edge machine learning techniques.
• Analysis of models, architectures, and hardware requirements for resource-constrained devices.
• Case study of a microcontroller-based edge machine learning implementation.

Main Results:

• Identification of key methods for deploying ML on edge devices.
• Assessment of current solutions and their limitations.
• Demonstration of a practical "Hello World" example for edge ML on microcontrollers.

Conclusions:

• Edge machine learning is crucial for overcoming IoT network limitations.
• Further research is needed to optimize ML models for low-power edge devices.
• This work lays the foundation for the development of the Internet of Conscious Things.