Künstliche Intelligenz auf Edge-Geräten

• Simon Franz Schneeberger

Nowadays, artificial intelligence (AI) is used in many different areas, including the private sector. Much of this is based on a central AI computing unit, the cloud. However, such a use is not possible in every application and therefore the interest in decentralized computing units for AI applications, the so-called edge devices, is increasing. Privacy issues in particular can be solved sensibly with the edge variant. This thesis investigates the potential and also deals with the evaluation of AI models on such devices.

The bachelor thesis gives on the one hand an introduction to the topic “Artificial Intelligence”, on the other hand different edge devices are being tested regarding performance. In addition, further criteria, such as documentation or assistance from the manufacturer itself, are included in the evaluation. The creation of an evaluation system for the different test devices and the subsequent evaluation are an essential part of this work.

The results of the tests show that edge devices are particularly interesting in terms of low energy consumption and high speed of data processing. In addition, the results of the evaluation criteria reflect the fact that development boards with weaker hardware can also handle AI computation and be considered for specific use cases. It is also shown that many pitfalls can make the execution of AI models more difficult. For this purpose, solution approaches are provided to facilitate the start. Furthermore, data preparation is an important component for an optimal performance of AI models on edge devices and should be considered from the beginning.