Bulletin of the American Physical Society
APS April Meeting 2019
Volume 64, Number 3
Saturday–Tuesday, April 13–16, 2019; Denver, Colorado
Session T13: Computational Physics and Algorithms I |
Hide Abstracts |
Sponsoring Units: DPF Room: Sheraton Plaza Court 2 |
Monday, April 15, 2019 3:30PM - 3:42PM |
T13.00001: The Pandora multi-algorithm approach to pattern recognition in MicroBooNE Andrew Smith Pattern recognition is an essential stage in the reconstruction of particle interactions in liquid argon time projection chamber detectors, which are used to study the properties of neutrinos. The novel multi-algorithm approach implemented in the Pandora software uses many tens of algorithms to gradually build up an image of the event and has been used successfully for pattern recognition in a number of particle physics experiments including MicroBooNE at Fermilab, IL, USA. This talk outlines the operational principles of MicroBooNE and the algorithm flows used by Pandora for reconstructing and identifying neutrino interactions in the dense cosmic ray environment recorded by the detector. |
Monday, April 15, 2019 3:42PM - 3:54PM |
T13.00002: Efficient Neutrino Oscillation Parameter Inference with a Gaussian Process-Accelerated Feldman-Cousins Algorithm Lingge Li, Nitish Nayak, Jianming Bian, Pierre Baldi Neutrinos are tiny sub-atomic particles that carry no electrical charge and interact with matter only through the weak nuclear force, which makes them extremely hard to detect. There are three distinct types of neutrinos, called "flavors": (νe, νμ, ντ), each of which can "oscillate" into the other with a detectable probability. Many experiments have been set-up to measure the parameters governing the oscillation probabilities accurately, with implications for the fundamental structure of the universe. Very often, this involves inferences from tiny samples of data which have complicated dependencies on multiple oscillation parameters simultaneously. This is typically carried out using the unified approach of Feldman and Cousins which is very computationally expensive, on the order of tens of millions of CPU hours. In this work, we propose an iterative method using Gaussian Process to efficiently find a confidence contour for the oscillation parameters and show that it produces the same results at a fraction of the computation cost. |
Monday, April 15, 2019 3:54PM - 4:06PM |
T13.00003: lichen and h5hep: moving toward a ROOT-less workflow for HEP analyses Matthew Bellis, Madeline Hagen High-Energy Physics (HEP) analyses are almost entirely performed using ROOT which contains everything from its own file format to HEP-specific classes to plotting utilities. This monolithic structure means that most internal aspects of ROOT are optimized by hiding much of the inner workings and by keeping users in this walled garden. Python has become a dominant interface or “glue” language for both HEP and non-HEP science and there are multiple python interfaces to ROOT. But these are just other portals to this walled garden. In order to move away from ROOT and interface more organically with the broader python ecosystem, we developed h5hep, a file format that builds upon the widely-used HDF5 and mimics the ROOT file functionality, and lichen, a set of tools that provide basic plotting and fitting utilities familiar to HEP analysts. Both modules make use of standard python libraries so installation is trivial. While these libraries have primarily been used for education and outreach we suggest that they could also be useful for data preservation, open data analyses, and perhaps bioinformatics. Speed, memory, and file size performance will be presented and compared to the analogous ROOT applications. |
Monday, April 15, 2019 4:06PM - 4:18PM |
T13.00004: Machine Learning Applied To Background Events Identification In LUX Dark Matter Experiment Samuel Chan In most LUX data analyses, the collaboration have mostly relied on the LUX Data Processing Framework’s output known as the "reduced quantities" (e.g. event energy, position etc.). However some info embedded in the unprocessed multichannel photomultiplier tube (PMT) time traces were lost in the processing. To extract these info and improve current existing analysis, a technique uses convolutional neural networks (CNN) in the discrimination between a single S2 vertex vs. a double S2 vertex was developed. After training the CNN with all 122 channels of PMT waveforms, it can correctly identify double S2s that are partially merged together and look like a single vertex but with substructure. Such merged vertex signal would arise from two simultaneous particle interactions that occur within a few mm. Implementing this new identification technique not only reduces the hand scanning effort required, but also improves our analyses in (1) eliminating backgrounds from conventional source, and (2) Identifying rare event signals. The preliminary results of applying this technique to different physics searches such as xenon isotopes double decay half life estimate will be presented. |
Monday, April 15, 2019 4:18PM - 4:30PM |
T13.00005: Deep Learning for Neutrino Physics on NOvA: Successes and Lessons Fernanda Psihas NOvA is a long baseline neutrino experiment which measures muon-flavor to electron-flavor oscillations from neutrinos and anti-neutrinos produced in the NuMI beam at Fermilab. Over the past few years, we have adapted techniques from the field of computer vision to fundamental parts of NOvA's analysis such as signal selection, final state identification, and energy reconstruction. The adaptation of deep learning algorithms into analysis tools has been fruitful, and is becoming more widespread within NOvA and in the field. In this overview of NOvA's deep learning program I will showcase our applications to detector data analysis, as well as the improvements in selection efficiency obtained for anti-neutrino events in our latest result. I will also highlight ongoing development of new applications for a varied range of reconstruction tasks. I will discuss challenges associated with adapting these techniques to detector data, and lessons learned from our uses of deep learning in combination with NOvA's traditional reconstruction methods. |
Monday, April 15, 2019 4:30PM - 4:42PM |
T13.00006: Machine learning for event classification and automated discovery of new physics at the LHC experiments Sergei Chekanov A method for an automated preparation of the feature space for various supervised artificial neural networks (ANN) used for searches for new physics at the LHC is presented. The proposed standardization of the ANN inputs allows "fingerprinting" of final state of collision events, translating experimental data from particle colliders to the language convenient for machine learning techniques. This can simplify searches for experimental signatures of new physics at the LHC. The method was illustrated using Monte Carlo event generators for several models beyond the Standard Model. The Monte Carlo simulations were also used to illustrate the usability of this approach for general event classification problems. |
Monday, April 15, 2019 4:42PM - 4:54PM |
T13.00007: Measurement of the WW Cross Section using the Random Forest Algorithm Thoth Gunter The pp→WW process has historically shows discrepancies between predicted and measured cross-section values, greater than other electro weak processes. Standard analyses rely heavily on jet multiplicity cuts to remove major backgrounds. These cuts lead to hard to calculate log correction factors which could be the cause of the discrepancies. We present the measurement of the WW→2l2nu cross-section without strict jet multiplicity cuts, relying instead on a set of Random Forest Classifiers. With this technique we measure the pp→WW→2l2nu cross-section. |
Monday, April 15, 2019 4:54PM - 5:06PM |
T13.00008: Application of Machine Learning Techniques to study Jet Substructure Cristina Mantilla Suárez High Lorentz boosts pose a challenge to the reconstruction of hadronically decaying heavy particles (top, W, Z, H) as their decay products are collimated. An efficient identification of hadronically decaying heavy particles increases the sensitivity in searches for heavy new (beyond the standard model) particles and opens the high momentum phase space for standard model measurements of the top quark, W, Z and H. Machine learning for heavy flavor jet-tagging have been increasingly explored. In this talk, we present the machine learning based heavy-tagging algorithms studied in the CMS experiment at 13 TeV. The performance of these algorithms is studied in simulation. |
Monday, April 15, 2019 5:06PM - 5:18PM |
T13.00009: Quark-Gluon Discrimination at the Large Hadron Collider Owen W Tower, Amitabh Lath, Abhijith Gandrakota, Kevin Nash, Duncan Adams Quarks and gluons are part of the Standard Model of particle physics, but cannot be observed directly in high energy physics experiments since they appear as a shower of hadrons called jets. Analyses of jets that includes their substructure is a recent development that holds the promise of being able to differentiate between gluon jets and quark jets. Since quark jets are the main focus for many new physics searches, reduction of background from gluon jets increases the sensitivity to new physics signatures. In order to distinguish between quark and gluon jets, we create a Quark-Gluon Likelihood (QGL) discriminant and apply it to jets from simulated new physics signals as well as background. The QGL discriminator uses neural networks, which take input variables such as the primary vertices, transverse momenta of jets, and other parameters related to the jet substructure to calculate the QGL discriminator values. We show preliminary results showing improvement in signal sensitivity obtained from simulated new physics samples of supersymmetric gluino and SM background consisting of multiple jets. |
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