End-to-end particle and event identification for regular and boosted topologies with CMS Open Data

21 May 2019, 16:40
20m
Matagorda (Omni Hotel)

Matagorda

Omni Hotel

900 N Shoreline Blvd, Corpus Christi, TX 78401
Oral Machine Learning, Big Data and Quantum Information Machine Learning, Big Data and Quantum Information

Speaker

Emanuele Usai (Brown University (US))

Description

From particle identification to the discovery of the Higgs boson, neural network algorithms have become an increasingly important tool for data analysis at the Large Hadron Collider. We present a novel approach to event and particle identification, called end-to-end deep learning, that combines deep learning image classification algorithms with low-level detector representation. Using two physics examples as references: quark and gluon discrimination and top quark jet tagging, we demonstrate the performance of the end-to-end approach using high-fidelity detector simulations from the CMS Open Data. Additionally, we explore the relevance of the information collected from various sub-detectors and describe how end-to-end techniques can be useful for full-event interpretation.

Primary authors

John Alison (Carnegie-Mellon University (US)) Sitong An (CERN, Carnegie Mellon University (US)) Michael Andrews (Carnegie-Mellon University (US)) Bjorn Burkle (Brown University (US)) Sergei Gleyzer (University of Florida (US)) Ulrich Heintz (Brown University (US)) Meenakshi Narain (Brown University (US)) Manfred Paulini (Carnegie-Mellon University (US)) Prof. Barnabas Poczos (Carnegie Mellon University) Emanuele Usai (Brown University (US))

Presentation materials