Speaker
Description
The derivation of indirect constraints on dark matter annihilation in our Galaxy from cosmic-ray antiprotons requires computationally expensive simulations of cosmic-ray propagation. I will present a new method based on Recurrent Neural Networks that significantly accelerates simulations of cosmic-ray antiproton spectra from secondaries and dark matter and achieves an excellent accuracy. Importance sampling is identified as particularly suitable for efficiently marginalizing over the nuisance parameters related to cosmic-ray propagation while ensuring that the networks are only evaluated in well-trained parameter regions. The method allows to investigate a wide range of dark matter models and it speeds up the runtime by at least two orders of magnitude compared to conventional approaches. I will illustrate our method on two examples: First, for a generic dark matter model annihilating into a pair of $b\bar b$-quarks and and, second, for the scalar singlet dark matter model.
