Bulletin of the American Physical Society
60th Annual Meeting of the APS Division of Plasma Physics
Volume 63, Number 11
Monday–Friday, November 5–9, 2018; Portland, Oregon
Session GM10: Mini-Conference on Machine Learning, Data Science, and Artificial Intelligence in Plasma Research III |
Hide Abstracts |
Chair: Zhehui (Jeph) Wang, Los Alamos National Laboratory Room: OCC C124 |
Tuesday, November 6, 2018 9:30AM - 9:55AM |
GM10.00001: Temporal-spatial sparse coding for X-ray image analysis and interpretation Oleksandr Iaroshenko, Garrett Kenyon, Zhehui Wang Digital X-ray cameras are experiencing an ``arms race'' towards ever higher resolutions and frame |
Tuesday, November 6, 2018 9:55AM - 10:15AM |
GM10.00002: Machine learning for direct spectral measurement inversion Mark Cianciosa, Kody Law, Elijah Henry Martin, Abdullah Zafar, David L Green It is often the case physical models exists to predict the observable outcomes given a set of plasma conditions (A → B). For diagnostic measurements, the observations are known but the underlying plasma conditions are not. In the absence of the reverse model, (B → A), inverse methods determine these unknown quantities by searching parameter space for a set of optimal input parameters. By minimizing the difference between known and model outcomes, inverse methods determine the most probable parameters given the observable evidence. However, searches in parameter space can be nondeterministic and computationally costly making them ill suited for real-time applications such as feedback control. Machine learning methods can significantly reduce this computational cost by producing a direct model of the inverse representation (B → A). Using a physical model, a training set can be produced by sampling a wide range of parameter space offline allowing rapid inversion online. This presentation will show the predictive capability of neural networks trained on synthetic data when applied to experimental observations. |
Tuesday, November 6, 2018 10:15AM - 10:40AM |
GM10.00003: Parameter inference and model calibration with deep jointly-informed neural networks Kelli D Humbird, Jayson Luc Peterson, Ryan McClarren, Jay David Salmonson, Joseph M Koning “Deep jointly-informed neural networks” (DJINN) is a novel, automated process for determining an appropriate deep feed-forward neural network architecture and weight initialization based on decision trees. The DJINN algorithm reduces many of the challenges associated with training deep neural networks on arbitrary datasets by automatically and efficiently determining an appropriate architecture and initialization that results in accurate surrogate models. Furthermore, DJINN is readily cast into an approximate Bayesian framework, resulting in accurate and scalable models that provide quantified uncertainties on predictions. We show how DJINN models trained on ensembles of expensive computer simulations can be calibrated with experimental data to infer likely values of unknown physical quantities, such as flux limiters and laser power multipliers. 1. K. Humbird et al, arXiv:1707.00784 (2017). 2. S. F. Khan et al, Physics of Plasmas 23, 042708 (2016).
|
Tuesday, November 6, 2018 10:40AM - 11:00AM |
GM10.00004: VITALS: Surrogate Models and Genetic Algorithms to Accelerate Transport Model Validation Pablo Rodriguez-Fernandez, Anne E White, Alexander J Creely, Martin J Greenwald, Nathan T Howard, Francesco Sciortino, John C Wright, Clemente Angioni, Jonathan Citrin, Emiliano Fable, Simon J Freethy, Gary M Staebler The Validation via Iterative Training of Active Learning Surrogates (VITALS) framework [1] exploits surrogate strategies and a genetic-algorithm-based optimizer to test whether a combination of plasma parameters exists such that experimental transport measurements are captured by a transport model within error bars. For the first time, additional measurable quantities, such as incremental electron thermal diffusivity, temperature and density fluctuation levels, cross-phase angles, and particle diffusion and convection coefficients can be used simultaneously along with transport fluxes to study model validation. Furthermore, any combination of plasma parameters can be scanned with minimal computational cost. VITALS has been used successfully to validate the TGLF quasilinear turbulent transport model in the Alcator C-Mod and ASDEX-Upgrade tokamaks. First results indicate that these machine learning algorithms are suitable and adaptable as a self-consistent, fast, and comprehensive validation methodology for plasma transport codes. |
Tuesday, November 6, 2018 11:00AM - 11:20AM |
GM10.00005: Application of genetic algorithms to multi-objective data analysis in plasma x-ray spectroscopy Roberto C Mancini X-ray spectroscopy is a powerful diagnostic of plasma conditions. The associated data analysis requires the solution of inversion problems whose complexity and difficulty depends on the type and number of pieces of data that have to be taken into account. Recent advances in x-ray instrumentation have enabled the observation of arrays of spectrally, spatially, and time resolved data that have been key to unfold the spatial structure of inertial confinement fusion (ICF) implosion cores, i.e. a tomographic reconstruction of temperature and density spatial profiles. The information is encoded in the atomic and radiation transport physics that determines the observed radiation from the plasma. The data analysis requires the simultaneous and self-consistent consideration of a large number of photon-energy resolved x-ray intensity distributions, and search and optimization in a multi-dimensional parameter space. We discuss the application of the Pareto genetic algorithm to the solution of this data analysis problem, and illustrate the results with data from ICF implosion experiments performed at the OMEGA laser facility. The case of illustration is specific but the ideas and methodology developed are of general application. |
Tuesday, November 6, 2018 11:20AM - 12:20PM |
GM10.00006: PANEL DISCUSSION Michael Dikovsky, Robert Granetz, Alessandro Pau, Brian Sammuli, David Smith, Brian Spears Panel Discussion: Michael Dikovsky (Google Inc.), Robert Granetz (MIT), Alessandro Pau (University of Cagliari), Brian Sammuli (General Atomics), David Smith (U. Wisconsin), and Brian Spears (LLNL). |
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. |
© 2025 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