Machine Learning techniques for cross-talk rejection in two-neutron decays for MoNA

The study of nuclei close to the drip lines has lead to discovery of neutron halo, where the valence nucleon has an extended and diffuse matter distribution. The case of the two-neutron halos is particularlly interesting, since such systems can be interpreted as a closed core with two neutrons, forming a three-body system. Their study is fundamental to improve our understanding of both nuclear structure and dynamics [1, 2]. The Modular Neutron Array (MoNA) [3] is a highly efficient large-area neutron detector located at FRIB, and is ideal to study these sytems using fast beams. The study of two-neutron halos in modular neutron-arrays relies on the capacity to separate real coincidences from the ones produced by the re-scattering of the nucleons through the detector. This work focuses on the comparison of classic and machine learning algorithms used to select the real events. Traditionally, the use of casualty conditions yields a high rejection percentage, but as a counterback a reduction of detection efficiency is obtained. In order to have access to the most exotic nuclei where the beam intensities are critically low, both the precission and sensitivity of the chosen algorithm must be maximized. For this purpose a a set of DNN where trained, finding that its application can produce a significant increase of the aforementioned metrics