Event Description
Christina Peters, PhD
University of Delaware
Machine learning techniques have been successfully used throughout the sciences, including in the search for particle dark matter. However, these techniques often lack transparency and do not provide a physically interpretable model. Moreover, these techniques often only produce a single, locally-optimal solution to the parameter of interest, rather than a solution with uncertainty. Position reconstruction (estimation of the location of an interaction) within an astroparticle detector is often performed using machine learning techniques. Robust position reconstruction is paramount for enabling rare-event discoveries by dark matter detection experiments, as it allows for focus on interactions occurring only within the central volume of a detector where there are fewer backgrounds. In this talk, I will demonstrate a Bayesian network method for position reconstruction which is both physically interpretable and provides per-interaction uncertainties on position, using the XENONnT detector as a proof-of-concept. The XENONnT experiment was designed for the
direct detection of dark matter with a liquid xenon time projection chamber (TPC). Previous liquid xenon TPCs have set leading limits on WIMP dark matter, and this latest experiment in the XENON program features both increased target mass and decreased backgrounds. I will describe the problem of position reconstruction within particle dark matter detectors, introduce Bayesian networks — one of the two classes of probabilistic graphical models, and demonstrate the performance of the method for position reconstruction, using simulated data based on
the XENONnT detector.
|
