강의목표
- Understand FPGAs for AI acceleration :
1) Understand the workflow to go from Python to Hardware
2) Get an overview of the research & industry state on the subject
3) Know and understand the currently available tools
- Deploy an end-to-end solution on ZynQ
강의개요
- Theoretical courses (Reminders, Stakes, History, Tools used & Workflow presentation)
- Train, Quantize a neural network and retain its accuracy
- Prepare the neural network for HLS using brevitas and FINN
- Generate an IP based on the model using FINN (Verilog HDL based after HLS)
- Use the resulting IP in a ZynQ SoC to hardware accelerate the network
참고사항
♦ 출석 100%, 퀴즈 3/5문제 통과시 수료증이 발급됩니다.
♦ 수강신청 기간 내에 홈페이지에서 수강 취소해야 정상 취소처리 됩니다.
♦ 1개 교육에 대해 전일 결석시, 추후 3개월간 수강신청이 자동차단됩니다. (취소는 홈페이지에서 직접 가능)
♦ 외국인 강사이므로 영어로 진행되는 점 참고 부탁드립니다.
♦ 강의 교재는 별도로 제공되지 않고 강의 시 화면으로 보여집니다.
강좌상세
| 일자 |
2024-09-23 |
시간 |
13:00 ~ 14:00 |
강사 |
BABIN-RIBY Hugo 선임연구원 Duncha |
| 내용 |
○ Reminders : Neural networks basics, MNIST
○ Quantization and benefits for FPGA, data types
○ FPGA acceleration, real applications, how are we going to use Zynq |
| 일자 |
2024-09-23 |
시간 |
14:00 ~ 16:00 |
강사 |
BABIN-RIBY Hugo 선임연구원 Duncha |
| 내용 |
○ Describing the used workflow and tools :
- PyTorch, Brevitas, ONNX
- HLS principles & tools + small hands-on example
- FINN presentation, workflow, current state of research
- ZynQ workflow, Programmable Logic / Processing system integration
- Communication protocol : AXI4
○ How to create and use an IP, How will we apply this to our project |
| 일자 |
2024-09-24 |
시간 |
13:00 ~ 15:00 |
강사 |
BABIN-RIBY Hugo 선임연구원 Duncha |
| 내용 |
○ Create a docker container environment with FINN, brevitas, etc… (Students
will need Vivado/Vitis/VITIS HLS 2023.2 on their system)
○ Sketch a simple MNIST model, train it, Do the same with Quantization,
compare and be critical.
○ Experiment with the model (get used to the tools, see how data types and
layers affect training time and accuracy) |
| 일자 |
2024-09-24 |
시간 |
15:00 ~ 17:00 |
강사 |
BABIN-RIBY Hugo 선임연구원 Duncha |
| 내용 |
○ Prepare the network for FINN
○ Use FINN to create an IP, modify requirements and compare ressources /
performances.
○ Export & Examine the resulting IP, take the time to understand how it works
and how it interfaces with the processing system. |
| 일자 |
2024-09-25 |
시간 |
13:00 ~ 15:00 |
강사 |
BABIN-RIBY Hugo 선임연구원 Duncha |
| 내용 |
○ Integrate the IP in Vivado, export hardware
○ Create software in Vitis
○ Run on FPGA, troubleshooting, experimenting & compare results with original
model |
강의장소
온라인 진행 (ZOOM)
담당자 연락처
|
