Precise and fast model prediction with machine learning



With this MDSI project we explore neural networks (NN) and other machine learning (ML) techniques for fast and precise model predictions. The prime application in this project will be the KATRIN experiment, which performs a high-precision spectroscopic measurement of the tritium beta decay spectrum to directly probe the neutrino mass with unprecedented sensitivity. The bottleneck in the KATRIN analysis is the computationally intensive analytical calculation of the integral tritium spectrum. A precise and fast prediction of the spectrum via a ML model would open up completely new possibilities for the KATRIN data analysis. Beyond that, we will use the KATRIN case to perform a theoretical analysis of high-precision ML models and we will explore the application of the developed techniques in other high-precision experiments.