Nick Prouse (TRIUMF)

Enhancing water Cherenkov event reconstruction with machine learning for Hyper-Kamiokande

Machine learning has the potential to enhance the sensitivities of water Cherenkov detectors by improving event reconstruction, suppressing backgrounds and reducing systematic uncertainties. These improvements will be vital in achieving the precision measurements that current and next-generation detectors are now aiming to perform. Inspired by recent breakthroughs in computer vision, the use of deep-learning techniques like convolutional neural networks and related architectures has begun to surpass the performance of traditional methods in broad areas of physics data analysis. These techniques are well suited to the data of imaging detectors like the water Cherenkov detectors, particularly to exploit additional spatial and directional information from higher granularity PMTs developed for the Hyper-Kamiokande experiment. In this talk I will discuss the progress being made towards particle classification and reconstruction, along with the challenges being addressed and future plans for other applications