A Multi-Task Video Transformer Model for Compact Imaging Atmospheric Cherenkov Telescopes

Imaging Atmospheric Cherenkov Telescopes (IACTs) are used to study the highest-energy gamma rays. While achieving great performance, their large scale leads to several issues that can be addressed by using a smaller telescope instead. With current analysis methods, such a small instrument would not perform well. In this work, we develop a new deep learning-based analysis that performs end-to-end video event reconstruction. We employ a Video-Vision Transformer Model with a factorized encoder that performs the three downstream analysis tasks. We show that we can improve the telescope's sensitivity by one order of magnitude compared to that achieved with standard methods. This work is the first application of a transformer-based analysis to IACTs and paves the way for time-domain gamma-ray astronomy, and opens the possibility of constructing compact, cost-effective yet sensitive instruments that could provide full-sky coverage and fast-transient alert follow-up for future instruments.