About
Trying to modulate the RF cavity of a Synchrotron Light Source by leveraging Reinforcement Learning, resulted in a hardware implementation of the Gated Recurrent Unit (GRU) on the Versal AI Engine, by AMD Xilinx and extremely efficient in performing the main numerical operations needed by the model.
RNNs have been designed to handle time series, and they are the perfect candidates to handle this kind of task. Although RNNs don't parallelize well, their ability to distill input and pass information into the hidden state, could be beneficial for real time control tasks. I will firstly introduce the AI Engine and its features. Second, I will introduce the GRU cell. Then I will show how numerical operations are being implemented in the AI Engine and its memory limitations. Finally, I will discuss how we tackled the lack of built-in activation functions, since the tanh and sigmoid function implementations are needed..
0:00:00 Slide 1
0:01:02 Slide 2
0:02:03 Slide 3
0:20:01 Slide 4
0:21:07 Slide 5
0:25:21 Slide 6
The talk
Credits and licence
- Institution
- CERN
- Department
- FPGA Developers' Forum (FDF)
- Speaker
- Michail Sapkas
- Year
- 2025
- Language
- English
- Source
- CERN
- Licence
- CERN, as CERN states it. CERN's terms
- Attribution
- © CERN. CERN provides its audiovisual media free of charge for educational and informational use, credited to CERN; this lecture is shown in CERN's own player.
- Who can watch
- CERN gives no rating. Invisible Network shows it to Teens profiles and older, judged from what the source says about it. How courses are placed