Image Classification Applied to High Energy Physics Events
Loading...
Date
2015-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
The Large Hadron Collider (LHC) is being upgraded to produce proton-proton collisions at 13TeV at a higher luminosity by the end of 2015. While increasing the chances for rare, interesting phenomena, the new environment will be substantially nosier, making it much more difficult for traditional analyses, which rely on largely isolated particles, to extract the associated signals. Deep Convolutional Neural Networks (DCNNs), computational models inspired by the visual cortex, have greatly surpassed the performance of other methods in image classification competitions. This project aims to evaluate the usefulness of DCNNs in the new environment at the LHC by literally viewing events as images. This paper describes how detector information can be converted into images and how a DCNN can be trained and optimized to distinguish tt and W+4jets, two well-understood types of events. Current results show a DCNN, with only calorimeter information, can achieve roughly equivalent performance to that of a traditional multivariate technique utilizing the full detector.
Description
2015 Undergraduate Research Forum for Engineers and Architects, 1st Place
2015 Natural and Mathematical Sciences Undergraduate Research Forum, Best Poster
2015 Natural and Mathematical Sciences Undergraduate Research Forum, Best Poster
Keywords
high energy physics, convolutional neural network, deep learning, multivariate analysis