Bulletin of the American Physical Society
APS April Meeting 2018
Volume 63, Number 4
Saturday–Tuesday, April 14–17, 2018; Columbus, Ohio
Session B08: Machine Learning, and Other Advanced Computational Techniques |
Hide Abstracts |
Sponsoring Units: DPF DNP Chair: Matthew Graham, SLAC National Accelerator Laboratory Room: A110 |
Saturday, April 14, 2018 10:45AM - 10:57AM |
B08.00001: Neutrino Identification using Convolutional Neural Networks in the NOvA Experiment Ryan Murphy Convolution Neural Networks (CNNs) have been successful in many complex computer vision problems in image identification and analysis due to recent advents in efficient GPU training. Due to these successes, effort was put forth to bring this technology to applications in HEP where the data could be easily converted into images. NOvA is a long baseline neutrino oscillation experiment designed to visually identify and reconstruct neutrino interactions in our detectors. Using a CNN, we created a algorithm for the identification of neutrino interaction types and neutrino flavors using the Caffe framework. In 2016, NOvA released the first HEP result to employ a CNN in the $\nu_{\mu}$ to $\nu_{e}$ oscillation channel. In this talk I will describe our implementation of CNN in the Caffe framework, it’s application to NOvA events, and improvements made to our original implementation. [Preview Abstract] |
Saturday, April 14, 2018 10:57AM - 11:09AM |
B08.00002: Electron Neutrino Energy Reconstruction with Convolutional Neural Network Shiqi Yu NOvA is a long baseline neutrino oscillation experiment. It uses two functionally identical liquid scintillator detectors to measure $\nu_{e}$ appearance and $\nu_{\mu}$ disappearance at the Far Detector in the $\nu_{\mu}$ beam produced by the NuMI facility at Fermilab. \\ NOvA uses a convolutional neural network(CVN) to identify neutrino events. A different network, called ``Prong-CVN'', has been used to classify reconstructed particles in each event as either lepton or hadron. Within each event, hits are clustered into prongs to reconstruct final state particles and these prongs form the input to this new classifier. Classified particle energies are then used as input to an electron neutrino energy estimator. Improving the resolution and systematic robustness of NOvA's energy estimator will improve the sensitivity of the oscillation measurements. In this talk, I will present our methods to identify particles with Prong-CVN and the following approach to estimate $\nu_{e}$ energy for signal events. [Preview Abstract] |
Saturday, April 14, 2018 11:09AM - 11:21AM |
B08.00003: Machine learning approaches to measure the hadronic recoil for a W mass precision measurement with the CMS experiment at LHC Olmo Cerri Given the large amount of data collected at LHC, an effective way of searching for new physics and test the SM consistency is comparing high precision theoretical prediction with observables measured in experiments. At the state of art, a better measurement of the W boson mass is of great importance in this procedure since the theoretical prediction uncertainty is much smaller than the world average measured one. In order to achieve the required relative precision of $10^{-4}$, it is crucial to master the detector, the analysis, and the theoretical predictions at an unprecedented level. An innovative study of the hadronic recoil produced in association with the W boson at LHC, which represent one of the main systematic uncertainty to the W mass measurement, is described in this work. Using a semi-parametric regression based on neural networks, a new and better experimental definition of this quantity is achieved. The power of the new definition is tested in terms of systematic uncertainties, which are evaluated in a new and original way. Thanks to the event-by-event prediction of the recoil pdf, the new definition results in a significative reduction of the related systematic uncertainty, estimated to be up to a factor 3 smaller. Full work link: http://cds.cern.ch/record/2285935 [Preview Abstract] |
Saturday, April 14, 2018 11:21AM - 11:33AM |
B08.00004: Measuring the WW cross section using the Random Forest MVA Thoth Gunter, Michael Schmitt The pp$\rightarrow$WW process is an electro weak process that has historically shown discrepancies between predicted and measured cross section values, far greater than the other electro weak processes. Standard WW cross section measurements rely on strict jet multiplicity cuts, introducing additional hard to calculate log correction factors. We present the measurement of the WW$\rightarrow$2l2nu cross section using the Random Forest MVA technique using 13 TeV proton-proton collision at the LHC recorded by the CMS detector. With this technique we measure the pp$\rightarrow$WW$\rightarrow$2l2nu cross section. [Preview Abstract] |
Saturday, April 14, 2018 11:33AM - 11:45AM |
B08.00005: Deep Learning the Jet Response Rohan Bhandari, Raghav Elayavalli, Alexx Perloff, Andrew Whitbeck Understanding and correcting the detector response to any observable of interest is an important task for experimentalists and is necessary for removing the impact of detector imperfections and facilitating direct comparisons with theoretical predictions. Determining the detector response to jets, however, is particularly difficult as jets are composed of many correlated particles of various types that are incident on a large area of a detector. To cope, current methods ignore the extra information of the jet’s internal structure contained by these particles and simply parameterize the response as a function of the jet’s transverse momenta and rapidity. The rise of “Deep Learning”, however, provides a framework in which to understand these correlations and extract robust measurements of the jet response. By representing jets as images, a deep convolutional neural network can be trained on lower-level features of the jet structure. These “jet images” allow the network to learn the jet response as a function of observables such as the jet fragmentation and energy distribution. We show that with jet images one can effectively reproduce the results of existing methods, while additionally exploiting the jet’s internal structure, leading to improved measurements of the jet response. [Preview Abstract] |
Saturday, April 14, 2018 11:45AM - 11:57AM |
B08.00006: Utilization of Machine Learning to Enhance Background Subtraction at E906/SeaQuest Marshall Scott The SeaQuest experiment utilized the 120 GeV Main Injector beam at Fermilab in p$+$p and p$+$d fixed target collisions to study the flavor asymmetry of the proton through the Drell-Yan process. Though the Drell-Yan process is clean, its cross section is tiny compared to the nuclear cross section. This coupled with the high intensity beam yields significant random background that must be removed. Monte Carlo simulations of Drell-Yan events from the beam dump and targets have been used to develop sets of analysis cuts. Machine learning, specifically Boosted Decision Trees and Probability Density Foams, have been used to augment these cuts and provide additional sets of cuts. A discussion of the merits of using these algorithms and a comparison with previous analysis cuts will be presented. [Preview Abstract] |
Saturday, April 14, 2018 11:57AM - 12:09PM |
B08.00007: A framework for PDFs at a photon collider Zack Sullivan We introduce a resolved photon PDF framework for photon colliders and set of photon PDFs. The energy of a photon-photon collider whose beams are created from gamma-electron backscatter will vary between ~0 and 400 GeV at a 500 GeV $e^-e^-$ collider. While fits to the probability to find a quark or gluon from a photon exist in CJK2 and GRV PDF sets, a photon collider requires the convolution of the probability to have a beam photon with a given energy with the probability to find the parton for each event. To accomplish this convolution during event production in a Monte Carlo generator is computationally prohibitively expensive (several minutes per integral). Hence, we create a table-based photon PDF with interpolation in an offline step, and provide an interface to LHAPDF to read in the results. This new general-purpose PDF makes possible fast phenomenological studies at photon-photon colliders, such as $b\bar b$ production, a major background to precision Higgs physics at a photon collider. [Preview Abstract] |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700