Speaker
Description
Lasers and accelerators are inherently complex systems, often requiring multi-input multi-output (MIMO) control strategies with demanding requirements on precision, speed, and scalability. As these systems push toward more stringent performance goals, traditional control techniques often face limitations in responsiveness and robustness.
In this talk, I will discuss how we’ve begun incorporating machine learning (ML) into feedback control loops to address some of these challenges. When integrated thoughtfully, ML models can provide fast, data-driven predictions and decisions that enhance control performance, particularly in complex environments.
I will highlight several examples where ML has contributed to improved outcomes. At LBNL, ML-based feedback reduced the response time of complex laser combining systems by nearly an order of magnitude. On the BELLA Petawatt beamline, we performed the first experimental demonstration of ML-driven shot-to-shot laser pointing stabilization, addressing bandwidth limits in conventional control systems. We’ve also developed lightweight reinforcement learning algorithms for various control scenarios and begun implementing ML models on FPGAs for real-time MIMO control.
These efforts are still ongoing, but suggest that ML can be a valuable and practical addition to modern control systems—offering improved precision, adaptability, and speed in demanding laser and accelerator environments.
Funding Agency
Office of Science, Office of High Energy Physics, of the US Department of Energy, and the Laboratory Directed Research and Development Program of Lawrence Berkeley National Laboratory
| I have read and accept the Privacy Policy Statement | Yes |
|---|---|
| Please consider my poster for contributed oral presentation | Yes |
| Would you like to submit this poster in student poster session on Sunday (August 10th) | Yes |
