Mono-X signal characterisation from two-component DarkMatter using Neural Networks

Abstract: In the recent decade, neural networks have become one of the leading tools in the search for the elusive nature of dark matter and its interactions through the analysis of numerically simulated data. This project introduces a minimal dark sector consisting of three additional fields and studies it through two mono-X signatures,mono-jet and mono-Z. A convolutional neural network is then implemented to identify and characterise thedark matter components of the generated signals. Using the data from the mono-jet signature, the networkclassifies the signals by number of dark matter components and then determines their masses; for the mono-Z, the network is also able to identify the spin of single component dark matter signals.The classification task is performed with perfect accuracy for both mono-X signatures. The regression task isnotoriously more complex. However, by increasing the size of the layers, the network is able to characterisethe masses of the dark matter components present in the signal with errors below 5% of the considered massranges. For both single and two-component dark matter signals, the network performs this task with betteraccuracy when using data from the mono-Z signature.