Evaluating the practical limitations of TinyML: an experimental approach

Tiny Machine Learning (TinyML) is a novel research field that opens up the possibility of embedding local intelligence into frugal objects thus creating new opportunities for building "networks of collective intelligence". Energy consumption reduction is probably the main reason why TinyML deserves attention in terms of sustainability, but it is not the only one. The low costs of the hardware and the possibility to increase the level of data security and user privacy are outstanding reasons as well. In this work, we present the results of a sensitivity analysis we have conducted to evaluate the performance of a Random Forest with data collected by a state-of-the-art hardware device. We focused on assessing the sensitivity of the detection of sounds, colors, and vibrations patterns. Results show that TinyML can be absolutely used to properly discriminate among several ranges of sounds, colors, and vibrations patterns, paving the way for the development of new promising sustainable applications.