Deep learning algorithms applied to camera images of the MAGIC telescopes

MAGIC is a system of two ground-based Imaging Air Cherenkov Telescopes with a diameter of 17 meters, designed for the detection of very-high-energy gamma-rays. Its cameras are equipped with 1039 photomultiplier tubes each, providing a charge curve for every camera pixel. Integrated pixel charges and arrival times are extracted from these curves and combined to one camera image per event. Subsequent to the image cleaning, the image parameters are calculated to estimate the type of the incident particle as well as its direction and energy. Currently, this is achieved by individual methods. As an alternative, these tasks could be accomplished all at once, using machine learning algorithms on the uncleaned camera images which would render the image cleaning and the image parameter calculation redundant. A promising and novel approach in the field of astroparticle physics - especially suited for the task of image classification - is the application of deep learning algorithms (DLAs). They consist of multiple layers of neurons addressing different levels of data abstraction. The aim of this work is to obtain a DLA and compare its performance to that of the currently used methods. In this talk, the project of applying DLAs to camera images of MAGIC is introduced and the current status is presented.