Deep Learning Applications in Flavour Tagging

Abstract

Motivated by the application of data-driven solutions to the field of particle physics, in particular flavour tagging, we study the effectiveness of deep learning (DL) approaches for inclusive measurement within the Belle II environment and strangness tagging in the LHCb environment.
%
In the study, we compare the performance of an existing Boosted Decision Tree approach with a Bayesian neural network. In addition, we perform an in depth study on the selected features, investigating the signal inclusivity of DL models which gives insights into behaviours of the models.
%
We aim for classification speed and precision in the strange-quark jets tagging study. Therefore, we explore using a simple fully connected feedforward neural network to classify -jets among all light jet backgrounds. A comprehensive feature investigation is performed to understand the discriminating power of jet observable and the importance of particle identification.
%
Additionally, data-driven methodologies are also reshaping industrial practices. A study investigating the potential of DL in predicting realised volatility of a financial index is included. It is a collaborative project with Optiver where neural networks along with various training schemes are studied to maximise profits.