Karlsruhe 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 96: Data, AI, Computing 7 (uncertainties, likelihoods)

T 96.6: Talk

Thursday, March 7, 2024, 17:15–17:30, Geb. 30.33: MTI

Probabilistic Machine Learning for the XENONnT position reconstruction — •Sebastian Vetter — Karlsruhe Institute of Technology, Institute for Astroparticle Physics

The XENONnT detector is a dual-phase Xenon time projection chamber to search for Dark Matter. To fully exploit background reduction, it is important to know the exact position of events in the detector. The event position reconstruction is commonly performed by a combination of different neural networks (NNs). These NNs, like most machine learning models used in modern experiments, output a singular point in the parameter space. The parameter space in this example is the horizontal plane of the detector.

In this talk I will present and compare two ways of modifying NNs to change their output from a singular point to a probability density.

The resulting probability density functions provide information about the uncertainty of the predictions. The numerical value of the uncertainty can be used to filter for potentially incorrectly reconstructed events. The shape of the uncertainty distribution can be analyzed to learn about trends and biases in the position reconstruction, ultimately leading to an improved signal to background discrimination.

This work is supported in part through the Helmholtz Initiative and Networking Fund (grant agreement no. W2/W3-118). In addition, support by the graduate school KSETA at KIT is gratefully acknowledged.

Keywords: Dark Matter; Machine Learning