| Speaker: | Omar Alterkait (Tufts University) |
|---|---|
| Title: | Equivariant Networks and Differentiable Simulations: Innovative Approaches to Particle Trajectory Reconstruction and Detector Calibration |
| Date (JST): | Mon, Mar 25, 2024, 11:00 - 12:00 |
| Place: | Seminar Room A |
| Abstract: |
Particle detectors are essential for advancing particle physics and astrophysics. However, analyzing particle trajectories and optimizing detector performance pose challenges due to data complexity and limitations of traditional methods. We present two novel machine learning approaches to address these challenges. >>> Firstly, we introduce Euclidean Equivariant Neural Networks for particle trajectory reconstruction in liquid argon time projection chambers. By employing tensor products and spherical harmonics, we achieve Euclidean Equivariance, reducing reliance on data augmentation and improving efficiency. >>> Secondly, we propose an alternative approach to traditional calibration and event reconstruction in water Cherenkov detectors using an analytical differentiable simulation model. This enables simultaneous optimization of multiple parameters through gradient descent, addressing limitations of sequential calibration tasks. >>> These approaches showcase the synergy between physics and machine learning, with physics informing equivariant neural networks and machine learning enhancing detector calibration and reconstruction. Further research has the potential to advance particle physics and astrophysics understanding. |
| Remarks: | CD3 x APEC Special Seminar |
