Gamma/Hadron Separation in Imaging Air Cherenkov Telescopes Using Deep Learning Libraries TensorFlow and PyTorch

Abstract

In this work we compare two open source machine learning libraries, PyTorch and TensorFlow, as software platforms for rejecting hadron background events detected by imaging air Cherenkov telescopes (IACTs). Monte Carlo simulation for the TAIGA-IACT telescope is used to estimate background rejection quality. A wide variety of neural network algorithms provided by both libraries can easily be tested on various types of data, which is useful for various imaging air Cherenkov experiments. The work is a component of the Astroparticle.online project, which collaborates with the TAIGA and KASCADE experiments and welcomes any astroparticle experiment to join.

Multi-messenger Astroparticle Physics for the Public via the astroparticle.online Project

“…3. This application is based on results of the study [24]. The Astroparticle CNN Client microservice is implemented as an interactive service that provides access to online analysis for selecting gamma events against the background of hadronic events using the convolutional neural networks developed within this project.…”

German-Russian Astroparticle Data Life Cycle Initiative to foster Big Data Infrastructure for Multi-Messenger Astronomy

“…Training datasets were obtained using the CORSIKA program [22]. Software implementation was developed using the frameworks TensorFlow [23][24][25], PyTorch [23,26], and sklearn [26].…”

“…In the study [23], the primary particle for TAIGA IACT data was identified using convolutional neural networks (CNN). The recognition quality parameter of the developed CNN equals = 2.7-3.0, which is significantly higher than = 1.7 provided by the classification methods based on analysis of Hillas parameters.…”

The Preliminary Results on Analysis of TAIGA-IACT Images Using Convolutional Neural Networks

“…It is known that other IACT installations [5,6] have shown promising results in image analysis of model data using convolutional neural networks. CNNs were also used for the TAIGA-IACT model data of one telescope [7,8]. However an imbalance of particle fluxes is observed in an experiment, thus it is necessary to consider the classification of events in the case of an unequal ratio of gamma quanta and hadrons.…”