Implementation of an Optimal Statistical Inference to Reduce Systematic Uncertainties in the Higgs to Tau Tau Analysis at the CMS Experiment

The field of High-energy physics has seen substantial progress in its methods of data analysis, advancing with the increased possibilities of machine learning techniques and improving the accuracy of the obtained physics results, mainly through the application of neural networks. As High-energy physics experiments continue their measurements, the amount of available data is continually increasing. This increase will lead to the reduction of statistical uncertainties in many analyses and emphasize the importance of adequately addressing the then increasingly dominant systematic uncertainties. This work presents an advanced strategy to minimize statistical and systematic uncertainties through a neural network, improving upon the currently proposed implementation and extending the method from binary tasks with one class to tasks with multiple classes comprising multiple processes. Comparing to the results given by the usage of the cross entropy function for event classification, as a current baseline, an improvement in the statistical and overall uncertainty can be achieved by utilizing this strategy, which is demonstrated by the application on a reduced data set of the CMS experiment for the machine learning based Standard Model analysis of the Higgs boson decay into two tau leptons. The utilization of the strategy results in reduced uncertainty in a combined and differential measurement, enabled by the introduced extension to multiple classes.