Speaker
Description
Deep learning techniques have continued to evolve and find novel applications across scientific disciplines, and Graph Neural Networks (GNNs) have emerged as a high-performance architecture particularly suited for datasets with irregular topology. The MAGIC Telescope, comprising a pair of 17 m Imaging Atmospheric Cherenkov Telescopes (IACTs) located at Roque de Los Muchachos Observatory in La Palma, Spain, is designed to detect gamma rays from around 50 GeV to over 50 TeV. IACT arrays rely on a multilayered pipeline in which each particle registered by the detectors creates a stereo signal which must be calibrated, converted to an image, cleaned, parameterized, and ultimately labeled by several machine learning algorithms. In recent years, Convolutional Neural Networks (CNNs) have shown great promise in performing both classification and regression tasks, demonstrating a comparable performance on calibrated IACT images. In contrast, this study leverages raw data, consisting of a 30 ns waveform signal in each pixel and covering the entire camera. Due to the unconventional geometry of the MAGIC cameras and uneven time-slicing across pixels, we represent raw-level MAGIC data as a point cloud graph and employ GNNs for the classification algorithm for the first time in an IACT. Our preliminary trials indicate that GNNs not only represent a robust method for analyzing raw MAGIC data, but also show potential for fast on-site data reduction and even direct telescope triggering.
| Collaboration(s) | MAGIC |
|---|