Machine learning status and prospects in KamLAND-Zen

KamLAND-Zen is a neutrinoless double-beta (0nbb) decay search experiment using a large liquid scintillator detector. The energy resolution is extremely important for reducing the background of two neutrino double-beta decay, and the vertex resolution is necessary for reducing the gamma decay background from radioactive impurities originating from the detector components such as an inner balloon. To improve these, the development of a reconstruction tool using a Graph Neural Network is underway. Furthermore, the spallation of double-beta decay isotope: xenon nuclei by muons generates long-lived unstable nuclei, which has become a major background in the current search for 0nbb decay. As it cannot be completely removed by the delayed coincidence measurement, particle identification between beta and gamma for removal is expected. For this reason, the development of PID using machine learning, including KamNet, is in progress.

In addition, a method to remove xenon nucleus spallation based on correlation information with neutrons and muons is also being developed, and a method using PointNet is being studied.

As a future prospect, tuning of the detector simulations by generative networks is proposed and being researched. In this presentation, I will comprehensively present these studies on machine learning in KamLAND-Zen.