Speaker
Description
Jefferson Laboratory (JLab) is home to the Continuous Electron Beam Accelerator Facility (CEBAF) and four experimental physics halls. JLab’s data science portfolio includes projects to advance research in nuclear physics, accelerator facilities, and engineering. With a specific focus on expanding capabilities in machine learning (ML)-based uncertainty quantification, design and control, and developing interpretability techniques. Examples of ML being used in JLab’s experimental halls and in collaborations with Oak Ridge National Lab (ORNL) and Fermilab will be shown. From JLab, an example of a production application using ML in an experimental hall resulting in 35% improvement in physics statistics due to significant improvement in track reconstruction efficiency, as well as an example of online detector calibration through high voltage control that is expected to decrease computation time when compared to traditional methods. Two applications developed in collaboration with JLab will be presented that implement uncertainty quantification using the new Deep Gaussian Process Approximation method: anomaly detection for the Spallation Neutron Source accelerator at ORNL, and a surrogate model for booster control at Fermilab.
| speaker affiliation | Jefferson Lab |
|---|
