Pulse-shape dicrimination with deep learning in CRESST

Abstract

Many unsolved mysteries in our universe such as galaxy rotation curves, mass distribution of clusters etc. can be reasonably explained by the concept of dark matter. WIMPs (Weakly Interacting Massive Particles) are the most favored dark matter candidate. In the quest to experimentally observe WIMP dark matter, CRESST is an outstanding experiment setting the best exclusion limits for low-mass WIMPs ever since. The analysis of the raw data observed by a particular cryogenic detector TUM40 used by CRESST is very challenging as two distinct pulse shapes are observed leading to a two-class classification problem. Neural networks and deep learning evolved to high potential tools in the field of data science. This work uses state-of-the-art deep learning techniques to investigate the two-class classification problem observed in TUM40 data.