Multi-View Deep Learning for Imaging Atmospheric Cherenkov Telescopes

Abstract

This research note concerns the application of deep-learning-based multi-view-imaging techniques to data from the H.E.S.S. Imaging Atmospheric Cherenkov Telescope array. We find that the earlier the fusion of layer information from different views takes place in the neural network, the better our model performs with this data. Our analysis shows that the point in the network where the information from the different views is combined is far more important for the model performance than the method used to combine the information.