Identifying muon rings in VERITAS data using convolutional neural networks trained on images classified with Muon Hunters 2
Authors:
Kevin Flanagan,
John Quinn,
Darryl Wright,
Hugh Dickinson,
Patrick Wilcox,
Michael Laraia,
Stephen Serjeant
Abstract:
Muons from extensive air showers appear as rings in images taken with imaging atmospheric Cherenkov telescopes, such as VERITAS. These muon-ring images are used for the calibration of the VERITAS telescopes, however the calibration accuracy can be improved with a more efficient muon-identification algorithm. Convolutional neural networks (CNNs) are used in many state-of-the-art image-recognition s…
▽ More
Muons from extensive air showers appear as rings in images taken with imaging atmospheric Cherenkov telescopes, such as VERITAS. These muon-ring images are used for the calibration of the VERITAS telescopes, however the calibration accuracy can be improved with a more efficient muon-identification algorithm. Convolutional neural networks (CNNs) are used in many state-of-the-art image-recognition systems and are ideal for muon image identification, once trained on a suitable dataset with labels for muon images. However, by training a CNN on a dataset labelled by existing algorithms, the performance of the CNN would be limited by the suboptimal muon-identification efficiency of the original algorithms. Muon Hunters 2 is a citizen science project that asks users to label grids of VERITAS telescope images, stating which images contain muon rings. Each image is labelled 10 times by independent volunteers, and the votes are aggregated and used to assign a `muon' or `non-muon' label to the corresponding image. An analysis was performed using an expert-labelled dataset in order to determine the optimal vote percentage cut-offs for assigning labels to each image for CNN training. This was optimised so as to identify as many muon images as possible while avoiding false positives. The performance of this model greatly improves on existing muon identification algorithms, identifying approximately 30 times the number of muon images identified by the current algorithm implemented in VEGAS (VERITAS Gamma-ray Analysis Suite), and roughly 2.5 times the number identified by the Hough transform method, along with significantly outperforming a CNN trained on VEGAS-labelled data.
△ Less
Submitted 17 August, 2021;
originally announced August 2021.
Real-Space Green's functions for Warm Dense Matter
Authors:
M. Laraia,
C. Hanson,
N. R. Shaffer,
D. Saumon,
D. P. Kilcrease,
C. E. Starrett
Abstract:
Accurate modeling of the electronic structure of warm dense matter is a challenging problem whose solution would allow a better understanding of material properties like equation of state, opacity, and conductivity, with resulting applications from astrophysics to fusion energy research. Here we explore the real-space Green's function method as a technique for solving the Kohn-Sham density functio…
▽ More
Accurate modeling of the electronic structure of warm dense matter is a challenging problem whose solution would allow a better understanding of material properties like equation of state, opacity, and conductivity, with resulting applications from astrophysics to fusion energy research. Here we explore the real-space Green's function method as a technique for solving the Kohn-Sham density functional theory equations under warm dense matter conditions. We find the method to be tractable and accurate throughout the density and temperature range of interest, in contrast to other approaches. Good agreement on equation of state is found when comparing to other methods, where they are thought to be accurate.
△ Less
Submitted 6 January, 2021;
originally announced January 2021.
