Bulletin of the American Physical Society
2017 Fall Meeting of the APS Division of Nuclear Physics
Volume 62, Number 11
Wednesday–Saturday, October 25–28, 2017; Pittsburgh, Pennsylvania
Session 1WA: Modern Machine Learning Methods in Data Analysis |
Hide Abstracts |
Chair: Curtis Meyer, Carnegie Mellon University Room: Marquis A |
Wednesday, October 25, 2017 9:00AM - 9:30AM |
1WA.00001: Machine learning at LHCb Invited Speaker: Mike Williams The use of machine learning (ML) has become ubiquitous at the LHCb experiment, producing sizable improvements in physics performance. I will discuss the use of ML in the real-time analysis/trigger system, including for event classification and reconstruction. I will also discuss the use of ML for particle identification, offline candidate selection, etc. A critical aspect of the use of ML at LHCb involves performing data-driven calibration/validation of the response of each algorithm, which will be discussed in the context of several examples. [Preview Abstract] |
Wednesday, October 25, 2017 9:30AM - 10:00AM |
1WA.00002: Deep Learning for in the search for New Particles Invited Speaker: Daniel Whiteson I will describe how the revolution in deep learning has enhanced the power of particle physics to discover new particles. [Preview Abstract] |
Wednesday, October 25, 2017 10:00AM - 10:30AM |
1WA.00003: Finding Neutrinos in LArTPCs using Convolutional Neural Networks Invited Speaker: Taritree Wongjirad Deep learning algorithms, which have emerged over the last decade, are opening up new ways to analyze data for many particle physics experiments. MicroBooNE, which is a neutrino experiment at Fermilab, has been exploring the use of such algorithms, in particular, convolutional neural networks (CNNS). CNNs are the state-of-the-art method for a large class of problems involving the analysis of images. This makes CNNs an attractive approach for MicroBooNE, whose detector, a liquid argon time projection chamber (LArTPC), produces high-resolution images of particle interactions. In this talk, I will discuss the ways CNNs can be applied to tasks like neutrino interaction detection and particle identification in MicroBooNE and LArTPCs. [Preview Abstract] |
Wednesday, October 25, 2017 10:30AM - 11:00AM |
1WA.00004: COFFEE BREAK
|
Wednesday, October 25, 2017 11:00AM - 11:30AM |
1WA.00005: CPP Detector Design Using MVA Invited Speaker: David Lawrence The Charged Pion Polarizability(CPP) experiment is approved to run in Hall-D at Jefferson Lab using the GlueX detector. CPP requires that $\pi^{+{}} \pi^{-{}}$ production events be distinguished from $\mu^{+{}}\mu^{-{}}$ to better than 99\% accuracy. This drives the design of a new MWPC-based detector capable of separating the $\pi$ events from the $\mu$ events. A multivariate analysis of simulated data was initially done to study the feasibility of a detector with this level of performance. More recently, the design parameters of the detector have been refined using a similar technique. Details on the initial study and how machine learning has contributed to the detector design will be presented. [Preview Abstract] |
Wednesday, October 25, 2017 11:30AM - 12:00PM |
1WA.00006: Weighing the Dark and Light in Cosmology with Machine Learning Invited Speaker: Hy Trac Galaxy clusters contain large amounts of cold dark matter, hot ionized gas, and tens to hundreds of visible galaxies. They are the largest gravitationally bound systems in the Universe and make excellent laboratories for studying cosmology and astrophysics. Historically, Fritz Zwicky postulated the existence of dark matter when he inferred the total mass of the nearby Coma Cluster from the motions of its galaxies and found it to be much larger than the visible mass. Nowadays, the abundance of clusters as a function of mass and time can be used to study structure formation and constrain cosmological parameters. Dynamical measurements of the motions of galaxies can be used to probe the entire mass distribution, but standard analyses yield unwanted high mass errors. First, we show that modern machine learning algorithms can improve mass measurements by more than a factor of two compared to using standard scaling relations. Support Distribution Machines are used to train and test on the entire distribution of galaxy velocities to maximally use available information. Second, we discuss how Deep Learning can be used to train on multi-wavelength images of galaxies and clusters and to predict the underlying total matter distribution. By applying machine learning to observations and simulations, we can map out the dark and light in the Universe. [Preview Abstract] |
Wednesday, October 25, 2017 12:00PM - 12:30PM |
1WA.00007: Bridging the Particle Physics and Big Data Worlds Invited Speaker: James Pivarski For decades, particle physicists have developed custom software because the scale and complexity of our problems were unique. In recent years, however, the "big data" industry has begun to tackle similar problems, and has developed some novel solutions. Incorporating scientific Python libraries, Spark, TensorFlow, and machine learning tools into the physics software stack can improve abstraction, reliability, and in some cases performance. Perhaps more importantly, it can free physicists to concentrate on domain-specific problems. Building bridges isn't always easy, however. Physics software and open-source software from industry differ in many incidental ways and a few fundamental ways. I will show work from the DIANA-HEP project to streamline data flow from ROOT to Numpy and Spark, to incorporate ideas of functional programming into histogram aggregation, and to develop real-time, query-style manipulations of particle data. [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