Speaker
Description
Telling apart Gamma and Cosmic rays is an important part of observing VHE sources. However, at low energies (10-100 TeV), this can often be challenging due to degeneracies in how the two show up on detectors such as LHAASO-KM2A. In this work, we leverage semi-supervised learning and explore if it is able to perform better than the traditional method (Q-Cut). Semi-Supervised Learning is a type of ML algorithms that utilise large amounts of unlabelled data and mix it in with a small subset of labelled data. Using this framework, we set up a Multi-Layer Perception Neural-Network (MLP-NN) and a Boosted-Decision Tree (BDT), showing that SSL methods work perform just as well as the traditional method. As a test, we apply the ML models to observe Crab Nebula and compare our results to those of the Q-Cut method.
