Tiny Machine Learning for Resource-Constrained Microcontrollers

Table 3

Summary of emerging techniques of TinyML.

Technique	Main features
Federated learning	Edge devices collaboratively train an ML model
	Improved data privacy and security
	E.g., TinyFedTL [89], centralized, and decentralized approaches
Transfer learning	Uses older ML models to generate a new one
	E.g., TinyTL [91]
On-device learning	Uses streaming data for training ML models in a microcontroller
	E.g., TinyOL [22] and NanoEdge AI Studio[7]
LPWAN	Long-range, low power, and low data rate
	E.g., LoRaWAN, Sigfox, NB-IoT, and LTE-M