Speaker
Description
Strongly coupled hidden sectors predict a family of dark matter particles
which reside in the dark sector. They can be generalized into two broad
groups: stable dark hadrons, which do not interact with baryonic
matter and hence cannot be detected, and unstable dark hadrons,
which decay back to Standard Model particles and therefore can be
reconstructed. Here, we introduce DRAGON (Deep
Reconstruction via Attention and
Graph Object Network), a
semi-supervised approach to reconstructing these short-lived hadrons,
which are intermediate particles in the decay chain of jets. We
demonstrate an average factor of 3 improvement in sensitivity compared
to baseline traditional methods, as well as an approach that
generalizes/extrapolates beyond the training regime, unlike regressors.
