Speakers
Mr
Evan Racah
(Lawrence Berkeley National Laboratory)
Wahid Bhimji
(Lawrence Berkeley National Lab. (US))
Description
High Energy Physics has made use of machine-learning approaches such as artificial neural networks for some time. Recently however there has been considerable development of these techniques outside the HEP community, particularly in deep neural networks for the purposes of image recognition. In this work, we describe a deep-learning analysis pipeline, developed at NERSC, capable of revealing meaningful physical content by transforming the raw data from particle physics experiments into a learned high-level representation using deep convolutional neural networks, including in an unsupervised mode where no input physics knowledge or training data is used.
We demonstrate this pipeline operating on raw data from current particle physics experiments, including the Daya Bay Neutrino Experiment. Furthermore we show how supervised convolutional deep neural networks can provide an effective classification filter with, in the Daya Bay case, greater than 97% accuracy across different classes of physics events, significantly better than other machine learning approaches. Such pipelines have significant applications for use in other experiments triggers, data quality monitoring or physics analyses.
Primary authors
Dr
Craig Tull
(Lawrence Berkeley National Laboratory)
Mr
Evan Racah
(Lawrence Berkeley National Laboratory)
Lisa Gerhardt
(LBNL)
Mr Prabhat
(Lawrence Berkeley National Laboratory)
Peter Sadowski
(University of California Irvine)
Dr
Sang-Yun Oh
(University of California, Santa Barbara)
Mr
Seyoon Ko
(Seoul National University)
Wahid Bhimji
(Lawrence Berkeley National Lab. (US))
