Hunting Dark Matter Signals at the LHC with neural networks

We study several simplified dark matter models and their signatures at the
LHC using Neural Networks. We focus on the usual monojet plus missing
transverse energy channel, but to train the algorithms we organize the data
in 2D histograms instead of event-by-event arrays. This results in a huge
performance boost to distinguish between SM only and SM plus new physics
signals. We found that Neural Network results do not change with luminosity,
if they are shown as a function of S/Sqrt[B], where S and B are the number of
signal and background events per histogram, respectively. To keep a broader
approach, we do not specify the simplified models coupling values. This
provides flexibility to the method, since testing a particular model is
straightforward, only the new physics monojet cross-section is needed.
Furthermore, we discuss the performance of the networks under wrong
assumptions. Finally, we propose multimodel classifiers to search and
identify new signals in a model independent way, for the next LHC run.