Deep learning in physics exemplified by the reconstruction of muon-neutrino events in IceCube

Recent advances, especially in image recognition, have shown the capabilities of deep learning. Deep neural networks can be extremely powerful and their usage is computationally cheap once the networks are trained. While the main bottleneck for deep neural networks in the traditional domain of image classification is the lack of sufficient labeled data, this usually does not apply to physics where millions of Monte Carlo simulations exist. At the IceCube Neutrino Observatory, the reconstruction of muon-neutrino events is one of the key challenges. Due to limited computational resources and the high data rate, only very basic and simplified reconstructions limited to a small subset of data can be run on-site at the South Pole. However, in order to perform online analysis and to issue real-time alerts, a fast and powerful reconstruction is necessary. In this talk I present how deep learning techniques such as those used in image recognition can be applied to IceCube waveforms in order to reconstruct muon-neutrino events. These methods can be generalized to other physics experiments.