Runtime calculation of CNNs

Environment

"Innovations for tomorrow's production, service and work" aims to improve production in medical technology. The thesis is situated in the context of a self-learning method, which automatically detects errors in the braiding pattern of cardiovascular implants (stents) and suggests optimized adaptation parameters based on this.

For the practical realization of such a system it is of advantage to be able to calculate the runtime of the system, since thus e.g. the hardware requirement can be adapted accordingly. Since the system is designed in the area of Machine Vision, CNNs are used among other things.

Objectives

• Investigate and describe the runtime of
• CNN architectures resp.
• The individual layer types,
• Practical comparison of the calculated and real runtime
• As well as examine the correspondence of the calculated and real runtime on the basis of further criteria such as hardware usage, programming language, etc.

Requirements

• Structured and analytical approach
• German and English business fluent in word and writing
• Basic mathematical understanding
• Practical programming experience is helpful (preferably in Python or C++)
• Previous knowledge in the area of Machine Learning or Deep Learning is helpful