SUSY Dimensionality Reduction

Abstract

Machine learning algorithms are a great ally in the search for newPhysics. One of the many goals of Experimental High Energy Physics(HEP) is to identify events that are evidence of Physics Beyond the Stan-dard Model (SM). Machine learning algorithms can be used to achievebetter distinction between signal and background events in data analysisfor the purposes of signal classification. Additionally, machine learningmay be used to reduce the dimensionality of a dataset. Roughly speak-ing, a dimensionality reduction algorithm can infer correlations betweenvariables and combine them into unified (reduced) variables, thereby de-creasing the number of variables necessary to represent our information.This is of particular interest to new physics because these “reduced di-mensions” may reveal information about underlying processes which arenot given to us explicitly by detector signals. In this project, we seekto reduce the dimensionality of a supersymmetry (SUSY) classificationdataset using machine learning to construct these reduced dimensions andto evaluate our information loss using a neural network classifier.