Swiss scientists are investigating the use of CNNs to perform the low-level data analysis of Cherenkov telescopes data [arXiv:1907.02428]. The standard γ-hadron separation based on binary decision trees (BDTs) was outperformed under specific conditions. This network was used to process Crab N ebula data of LS T-1 and again the CNNs outperformed the standard analysis. Other networks were trained to reconstruct the energy and origin direction of γ-ray events. CNNs outperformed the standard analysis also for energy reconstruction. Further developments are ongoing.

Efficiency curves comparing several CNNs to the standard BDT algorithm for the separation of hadrons and γ-rays. Higher curves are better.