-
Leomar Mateus Radke
Universidade Federal do Rio Grande do Sul (UFRGS) Programa de Pós-Graduação em Engenharia Elétrica (PPGEE) Porto Alegre, Rio Grande do Sul
-
Max Feldman
Universidade Federal do Rio Grande do Sul (UFRGS) Programa de Pós-Graduação em Engenharia Elétrica (PPGEE) Porto Alegre, Rio Grande do Sul
-
Ivan Müller
Universidade Federal do Rio Grande do Sul (UFRGS) Programa de Pós-Graduação em Engenharia Elétrica (PPGEE) Porto Alegre, Rio Grande do Sul
Keywords:
Optimization, TinyML, Anomaly Detection, Low-Power Wide Area Network
Abstract
The advancement of Internet of Things (IoT) applications in the context of low-power long-distance networks today is notorious. However, some weaknesses also appeared, such as the security of the transmitted data, bandwidth and battery life of the devices. This work presents an evaluation of optimized Tiny Machine Learning (TinyML) models. The benefits of having an optimized algorithm in a sensor device are evaluated, where the data inference is performed locally. The performance of each of the techniques will be evaluated, as well as the reduction capacity they promote. A case study is presented in a LoRa network, where a dataset is used to evaluate the energy performance of the model. The result was an approximate 6x drop in power consumption in the edge anomaly detection.
