Machine Learning Applied to High Energy Physics

Abstract

Machine learning algorithms have gained traction in a variety of fields throughout the last decade. This final degree project focuses on a bank problem and on a high-energy physics problem: searching for a rare Λ0b decay. Two different machine learning methods are used: Neural Networks and Boosted Trees, implemented in three different Phython libraries: TensorFlow and Keras, PyTorch and XGBoost. Using the AUC-ROC curve, the models between the three libraries are compared, and finally, models try to predict whether the Λ0b decay happens for a given data. Results for the bank problem shows nearly the same performance for TensorFlow and PyTorch, while XGBoost seems significantly better. For the high-energy problem XGBoost seems better, followed by TensorFlow and last PyTorch. However, predictions made on new data shows similar performance for XGBoost and PyTorch.