Implement Neural Nets on FPGA
Machine learning is everywhere these days, and Convolutional Neural Networks (CNNs) are one of the most prolific forms of AI on the current market. When trying to use CNNs for live image detection, companies currently have the choice between buying power-hogging GPUs and expensive dedicated ASICs. Quite a few of these companies would like to be able to choose the middle ground by having an FPGA implementation of their networks. That’s where a design house like Easics comes in.
Easics has developped a platform for automatic efficient implementations of Convolutional Neural Networks on FPGA. In order to showcase this platform, we want you to develop, test and analyze a demonstrator application on our platform by implementing an existing neural net (e.g. Mask-RCNN, GoogleNet/Inception… ). For this you’ll use and enhance our current high-level code-generation flow.
During this summerjob, your task will be to write Python and C++ to control and optimize both the data flow and control flow throughout the FPGA. You’ll weigh in on design decisions both for a specific net implementation and general code-generation flow.
Your profile is:
- A problem solver
- Demonstrated programming experience in C++ and Python
- Basic experience with Git
Estimated time duration is 6 weeks in july/august, depending on your availability.
Send your resume, motivation letter and availability to email@example.com