The classification and categorisation of Gamma-Ray Bursts with machine learning techniques for neutrino detection

TL;DR

This paper explores using machine learning, especially unsupervised techniques, to classify Gamma-Ray Bursts into subpopulations based on various features, aiming to identify groups with higher neutrino emission potential.

Contribution

It introduces an unsupervised machine learning approach to categorize GRBs using multiple features, potentially revealing new subpopulations linked to neutrino flux.

Findings

Identification of potential new GRB subpopulations

Use of diverse features like T90, hardness, and spectra

Uncovering hidden patterns in GRB data

Abstract

While Gamma-Ray Burst (GRBs) are clear and distinct observed events, every individual GRB is unique. In fact, GRBs are known for their variable behaviour, and BATSE was already able to discover two categories of GRB from the T90 distribution; the short and long GRBs. These two categories match up with the expected two types of GRB progenitors. Nowadays, more features have been found to be able to further distinguish them, such as the hardness ratio or the presence of supernovae. However, that does not mean that it is by any means simple to categorise individual GRBs. Furthermore, more GRB categories have been theorised as well, such as low-luminosity or X-ray-rich GRBs. These different types of GRBs also could indicate a different neutrino spectrum, with different types of GRBs more likely to emit higher amounts of neutrinos. We present an ongoing effort to use machine learning to categorise and classify GRBs, searching for subpopulations that could yield a larger neutrino flux. We specifically use unsupervised learning, as it allows hidden patterns and correlations to come to light. With the help of features such as the T90, hardness, fluence, SNR, spectral index and even the full light curve and spectra, different structures and categories of Gamma-Ray bursts can be found.