Analyzing γ-rays of the Galactic Center with Deep Learning

Abstract

We present a new method to interpret the ${\hat{I}}^3$-ray data of our inner Galaxy as measured by the Fermi Large Area Telescope (Fermi LAT). We train and test convolutional neural networks with simulated Fermi-LAT images based on models tuned to real data. We use this method to investigate the origin of an excess emission of GeV ${\hat{I}}^3$-rays seen in previous studies. Interpretations of this excess include ${\hat{I}}^3$ rays created by the annihilation of dark matter particles and ${\hat{I}}^3$ rays originating from a collection of unresolved point sources, such as millisecond pulsars. Our new method allows precise measurements of the contribution and properties of an unresolved population of ${\hat{I}}^3$-ray point sources in the interstellar diffuse emission model.

24 pages, 11 figures