HAGRID -- High Accuracy GRB Rapid Inference with Deep learning

Since their discoveries in 1967, Gamma-Ray Bursts (GRBs) continue to be one of the most researched objects in astrophysics. Multi-messenger observations are key to gaining a deeper understanding of these events. In order to facilitate such measurements, fast and accurate localization of the gamma-ray prompt emission is required. As traditional localization techniques are often time consuming or prone to significant systematic errors, here we present a novel method which can be applied on the POLAR-2 observatory. POLAR-2 is a dedicated GRB polarimeter, which will be launched towards the China Space Station (CSS) in 2025. The CSS provides POLAR-2 access to a GPU, which makes it possible and advantageous to run a Deep Learning model on it. In this work, we explore the possibility to identify GRBs in real time and to infer their location and spectra with deep learning models. Using POLAR simulations and data, a feasibility experiment was performed to implement this method on POLAR-2. Our results indicate that using this method, in combination with real time data downlinking capabilities, POLAR-2 will be able to provide accurate localization alerts within 2 minutes of the GRB onset.