The ZuSE-KI-AVF project is now looking for a student assistant to work on a framework for accelerating neural networks on a massively parallel vector processor architecture. The ZuSE-KI-AVF project is about accelerating AI algorithms in the automotive domain using the massively parallel vector processor architecture VPRO.
The CNN-Converter Framework analyzes existing neural networks and maps them to the massively parallel vector processor architecture VPRO. The tasks consist of implementing different layers of neural networks in C++ in the framework. The layers must also be analyzed for fixed point formats and memory transfers must be optimized. A simple neural network (YOLO-Lite) has already been mapped and will be used as a template.