Marco Wagner, Heilbronn University, Germany

Machine Learning (ML) and especially its application to cyber-physical systems is an uprising field of research. Many approaches on how to leverage the power of ML even in small devices have been published and applied in recent years, forming the field of TinyML. While TinyML has been promising several benefits such as cost-reduction, privacy and more, studies on which of these benefits are actually crucial when deciding to apply a TinyML approach to a real-world problem have been missing so far. The author of this paper argues, that without understanding the "why", the community of researchers and industrial parties may not understand the reasons of applying TinyML and hence may head into research and development directions not increasing its success. This work analysis the application of TinyML approaches and the reasons behind in recent years for three important fields of application: consumer electronics, manufacturing and automotive. It determines the distribution of TinyML applications in the named field and examines which of the bespoken benefits of TinyML were actually driving the decision to use it. Furthermore, this work investigates in cross connections between the benefits and hence points out the main combinations of benefits to the adopter.

TinyML, embedded systems, embedded ML, machine learning, consumer electronics, automation, automotive.