Employing deep learning for particle identification at the Large Hadron Collider

Benjamin Hooberman, University of Illinois at Urbana-Champaign

Usage Details

The Large Hadron Collider (LHC) at CERN, in Switzerland, is the world's most powerful particle accelerator. The LHC recreates the conditions of the universe one tenth of a nanosecond after the Big Bang, by colliding together protons traveling at 0.99999997 times the speed of light 40 million times every second. Each proton-proton collision creates up to several hundred particles that pass through one of four detectors situated at the LHC interaction points. Reconstructing the collisions requires identifying these particles using their signatures in the detector. Recent advances in machine learning and artificial intelligence, known as deep learning, have made it possible to apply learning networks to many kinds of problems. In particular, identifying particles from their energy deposition in calorimeter cells bears a strong resemblance to problems in machine vision, in which objects are reconstructed from intensity values in pixel arrays. This project focuses on applying deep learning techniques to identify and measure particles produced in LHC collisions and recorded by the ATLAS detector. The petascale resources and graphical processing units available at Blue Waters will allow training and optimization of neural nets beyond what has previously been achieved, allowing for detailed investigations of their behavior for both particle physics and general applications.