Searching for dark jets with displaced vertices using weakly supervised machine learning

Abstract: If "dark quarks" from a confining hidden sector are produced at the LHC, they will shower and hadronize to dark sector hadrons, which may decay back to Standard Model particles within the detector, possibly resulting in a collimated spray of particles resembling a QCD jet. In this work we address scenarios in which dark hadrons decay with a measurable small displacement, such that the relevant background is dominated by heavy-flavor jets. Since dark sector parameters are largely unconstrained, and the precise properties of a dark QCD-like theory are difficult to compute or simulate reliably in any case, model-independent, data-based searches for such scenarios are desirable. We explore a search strategy employing weakly supervised machine learning to search for anomalous jets with displaced vertices. The method is tested on several toy signals, demonstrating the feasibility of such a search. Our approach has potential to outperform simple cut-based methods in some cases and has the advantage of being more model-independent.