Capability of using the normalizing flows for extraction rare gamma events in the TAIGA experiment

TL;DR

This work addresses the challenge of detecting rare gamma events in TAIGA IACT data where gamma flux is much smaller than the cosmic-ray background. It adopts normalizing-flow based anomaly detection, framing the problem in two ways: treating hadrons as anomalies or gamma events as anomalies, and uses Hillas features as inputs. On TAIGA-model data, the method achieves a ROC AUC of up to 0.657 in the hadron-as-anomaly configuration, indicating potential but not yet practical performance for gamma isolation. The study suggests scaling up NF architectures and data, and exploring both two-class and one-class learning strategies to improve separation and enable practical gamma detection.

Abstract

The objective of this work is to develop a method for detecting rare gamma quanta against the background of charged particles in the fluxes from sources in the Universe with the help of the deep learning and normalizing flows based method designed for anomaly detection. It is shown that the suggested method has a potential for the gamma detection. The method was tested on model data from the TAIGA-IACT experiment. The obtained quantitative performance indicators are still inferior to other approaches, and therefore possible ways to improve the implementation of the method are proposed.

Figures (4)

• Figure 1: A general scheme of the gamma-hadron separation based on NF models.
• Figure 2: The results for the $\gamma$-event selection considering $\gamma$ as the anomalies. Top left: the loss function vs. epochs of the training. Top right: the distribution of the samples in the plane of the two Hillas parameters ($\alpha$-angle and width); the colors correspond to the probabilities of the samples in the latent space. Bottom left: the histogram of the gamma-hadron distribution by the magnitude of the log-likelihood; the chosen threshold for class separation is shown. Bottom right: distribution of data instances by classes and values of classification metrics.
• Figure 3: The results by considering hadrons as the anomalies and the $\gamma$-events as the normal ones. Top right: the distribution of the samples in the plane of the two Hillas parameters ($\alpha$-angle and width); the colors correspond to the probabilities of the samples in the latent space. Bottom left: the histogram of the gamma-hadron distribution by the magnitude of the log-likelihood; the chosen threshold for class separation is shown. Bottom right: distribution of data instances by classes and values of classification metrics.
• Figure 4: ROC curve for the case of hadrons as the anomalies; AUC=0.657.