Automatic Design of Artificial Neural Networks for Gamma-Ray Detection

Filipe Assuncao,Penousal Machado,Ruben Conceicao,Joao Correia,Mario Joao Martins Pimenta,Bernardo Tome,Nuno Lourenco

doi:10.1109/access.2019.2933947

Filipe Assuncao, Penousal Machado + Show 5 more

Open Access

https://doi.org/10.1109/access.2019.2933947

Copy DOI

Abstract

The goal of this work is to investigate the possibility of improving current gamma/hadron discrimination based on their shower patterns recorded on the ground. To this end we propose the use of Convolutional Neural Networks (CNNs) for their ability to distinguish patterns based on automatically designed features. In order to promote the creation of CNNs that properly uncover the hidden patterns in the data, and at same time avoid the burden of hand-crafting the topology and learning hyper-parameters we resort to NeuroEvolution; in particular we use Fast-DENSER++, a variant of Deep Evolutionary Network Structured Representation. The results show that the best CNN generated by Fast-DENSER++ improves by a factor of 2 when compared with the results reported by classic statistical approaches. Additionally, we experiment ensembling the 10 best generated CNNs, one from each of the evolutionary runs; the ensemble leads to an improvement by a factor of 2.3. These results show that it is possible to improve the gamma/hadron discrimination based on CNNs that are automatically generated and are trained with instances of the ground impact patterns.

Highlights

High-energy gamma-rays constitute one of the best probes to investigate extreme phenomena in the Universe, such gamma-rays arising from fast rotating neutron stars or supermassive black holes
FITNESS FUNCTION To assess the fitness of each individual, we evaluate the model in the test partition, and compute the true positive rate (TPR) and false positive rate (FPR) to build the Receiver Operating Characteristic (ROC) curve; we consider the positive class as the instances classified as a proton
The impact patterns are stored as matrices of signal, where each position keeps the energy detected in a specific water-Cherenkov detectors (WCDs)

Summary

INTRODUCTION

High-energy gamma-rays constitute one of the best probes to investigate extreme phenomena in the Universe, such gamma-rays arising from fast rotating neutron stars or supermassive black holes. The detection of this kind of astrophysical radiation, whose energies span from 10 GeV up to 100 TeV, can be done at lower energies by satellite bourne detectors. The patterns of the secondary particles at the ground remain to be explored, some studies have shown that this might have some gamma/hadron discrimination power In this manuscript, we intend to explore the difference in the patterns at the ground, between gamma and proton induced showers, recurring to Artificial Neural Networks (ANNs). The gains in performance represent an improvement by a factor of up to 2.48; this indicates that with the same grid of sensors we can perform twice better than other methods; on the other hand, it can lead to investment savings because a smaller grid of detectors can be used

GAMMA AND PROTON SIMULATION

EVOLUTION OF CONVOLUTIONAL NEURAL NETWORKS

CONCLUSIONS AND FUTURE WORK