Fusion Deep Learning
William Tang of Princeton University is a Principal Research Physicist at the Princeton Plasma Physics Laboratory for which he served as Chief Scientist (1997-2009) and Head, Theory Department (1990-2002). He is Lecturer with Rank & Title of Professor in the Plasma Physics Section of the Department of Astrophysical Sciences, Participating Faculty at the Center for Statistics and Machine Learning (CSML), and member of the Executive Board for the "Princeton Institute for Computational Science and Engineering (PICSciE)," which he helped establish and served as Associate Director (2003-2009). He received his PhD in Physics with his dissertation carried out at the University of California Lawrence Livermore National Laboratory within the Department of Applied Science founded by Edward Teller, and is a Fellow of the American Physical Society.
He is internationally recognized for expertise in the mathematical formalism as well as associated computational applications dealing with electromagnetic kinetic plasma behavior in complex geometries, and has well over 200 publications with an "h-index" or "impact factor" on Google Scholar Citations of 60 with over 14,300 citations that include 40 papers each with at least 100 citations. He was U.S. PI for the G8 Research Council's "Exascale Computing for Global Scale Issues" Project in Fusion Energy - an NSF-funded international HPC collaboration (2011-2014).
Prof. Tang has made original contributions in basic plasma physics, fusion energy science, interdisciplinary computational science, and artificial intelligence/deep learning -- with associated awards including:
- the 2005 Chinese Institute of Engineers-USA Distinguished Achievement Award "for outstanding leadership in fusion research and contributions to fundamentals of plasma science."
- the 2019 Global Impact Award from NVIDIA Corporation for "groundbreaking work in using GPU-accelerated computing to unleash deep learning neural networks for dramatically increasing the accuracy and speed in predicting disruptions in fusion systems."
- the 2013 High Performance Computing Innovation Excellence Award from International Data Corporation for "using high-end supercomputing resources to carry out advanced simulations for the first time of confinement physics in large-scale magnetic fusion energy plasmas with unprecedented phase-space resolution and long temporal duration to deliver important new scientific insights."
Dr. Tang is the current leader of an AI/Deep Learning project that has produced a high-profile article in NATURE (April 2019) on "Predicting Disruptive Instabilities in Controlled Fusion Plasmas through Deep Learning" (J. Kates-Harbeck, A. Svyatkovskiy, W. Tang) that brings together the emergence of the exciting growth area of deep learning with an exemplar from the grand challenge area of clean energy via magnetic fusion. This work uniquely demonstrated for the first time that predictive software trained on one experimental device (e.g., the DIII-D tokamak in CA) to make accurate predictions on another (the larger and more powerful EUROFUSION JET tokamak in the UK). Moreover, it was the first FES DL code to efficiently utilize leadership class supercomputers (e.g., Titan, Summit in US; & Tsubame-3 in Japan) -- highlighted in his invited Keynote on “Deep Learning Acceleration of Progress toward Delivery of Fusion Energy" at the 2018 International Supercomputing Conference in Frankfurt, Germany. He is currently PI for prominent FES HPC Supercomputing discovery science projects including the ANL Early Science Project (ECP) on AURORA -- the first targeted DOE-SC Exascale System in 2021, the new “NESAP for Learning (N4L)” Project on the Perlumutter System at LBNL, and (as Co-PI with Zhihong Lin of UC Irvine) on the internationally top-rated SUMMIT System at ORNL.
Julian Kates-Harbeck grew up in Munich, Germany. At Stanford University he received his bachelor’s degree in physics and later earned a master’s in computer science (with a focus on AI and machine learning) there. Before entering graduate school, he co-founded a tech startup, where he was responsible for product, hiring, and strategy. He completed his PhD in physics at Harvard University, where he studied dynamics on complex social and biological networks, as well as applications of deep learning to fusion energy science. For his PhD studies, he was awarded the National Science Foundation GRFP, Department of Defense NDSEG, and Department of Energy CSGF fellowships.
Julian has now joined Kernel, a company aiming to build next-generation brain-machine interfaces. He has written about machine learning, biophysics, high-performance computing, and plasma physics.
Alexey Svyatkovskiy is a Senior Data Scientist at Microsoft, Cloud & AI. His research interests focus on machine learning algorithms that understand and generate source code, aiming to improve automated software developer tools in Azure cloud and Visual Studio Code IDE.
As a member of the FRNN team he contributes to the development of the prediction algorithm. He has ported the software to several HPC systems and collected performance results.
Alexey holds a PhD in Physics from Purdue University. His thesis research focused on dimuon physics analyses and charge particle reconstruction algorithms at very high energies. Dr. Svyatkovskiy has contributed to the activities of the CMS collaboration at the Large Hadron Collider towards the world-famous Higgs boson discovery, which was published in Physics Letters B in 2012.
Eliot Feibush leads the Princeton Visualization Consortium which includes the Princeton Institute for Computational Science and Engineering, the Geophysical Fluid Dynamics Laboratory, and PPPL. He joined the Computational Plasma Physics Group in 2001. Eliot received his Bacheolor of Architecture and Master of Science in computer graphics from Cornell University.
Eliot specializes in scientific visualization. As a member of the High Performance Computing Team, he is creating visualizations of fusion simulations and plasma physics experiments. Other efforts include developing scientific graphics software for showing data acquired from experiments and subsequent analysis.
Kyle Felker is currently a postdoctoral appointee at Argonne National Laboratory in the Leadership Computing Facility. He is the project’s appointee in the Aurora Early Science Program, where he is charged with preparing the group’s software for deployment on Aurora, the nation’s first exascale system planned for delivery in 2021.
Kyle has extensive experience in high-performance computing, software engineering, and computational physics, and he is developing as a practitioner of machine learning. He received his PhD in Applied and Computational Mathematics from Princeton University in 2019 and his BA from the University of Chicago in 2013.
Dan is a staff research physicist at Princeton Plasma Physics Laboratory in the Advanced Scenarios and Control group for the National Spherical Torus Experiment Upgrade (NSTX-U). His research focuses on optimization and feedback control for fusion reactor operation. His publications include work on control algorithm development for NSTX-U and DIII-D, and development of machine learning accelerated models for predicting fusion reactor performance and operational boundaries fast enough for real-time applications.
Keith Erickson is the Realtime Linux Engineering lead at the DOE Princeton Plasma Physics Lab (PPPL) managed by Princeton University. After designing real-time control systems while employed by Lockheed Martin, he now works on plasma control and protection systems for fusion energy magnetic confinement "tokamak" experiments both domestically and internationally. He is a leading expert in the design and implementation of real-time systems that achieve determinism thresholds of one microsecond or less through the innovative use of new technologies.
Dr. Ge Dong is a postdoctoral research associate at the Princeton Plasma Physics Laboratory. With her expertise in AI/Deep Learning she is investigating plasma disruption and instabilities using the modern Fusion Recurrent Neural Network (FRNN) software. With complementary expertise in both first-principles analytic theory and advanced gyrokinetic simulations she is using the state-of-the-art nonlinear electromagnetic Gyrokinetic Toroidal Code (GTC).
Dr. Bei Wang is a senior research software engineer in the Research Computing department at Princeton University. Currently, she is a member of IRIS-HEP and working with physicists to speed up the particle tracking algorithms on GPU and FPGA. She is also the co-PI of the Intel Parallel Computing Center (IPCC) at Princeton University. She received her Ph.D. in applied science with designed emphasis on computational science and engineering from the University of California at Davis.
Mitchell Clement completed his Ph.D. in feedback control of MHD instabilities on DIII-D while attending UC San Diego. Following graduation, he joined the Applied Physics and Mathematics department at Columbia University, where he continued Resistive Wall Mode feedback research at DIII-D and completed improvements to the DIII-D plasma control system. Following Columbia University, he joined PPPL as a postdoctoral researcher at DIII-D, developing deep learning models of Neoclassical Toroidal Viscosity torque and developing schemes for rotation profile control.
Princeton Plasma Physics Laboratory
100 Stellarator Road
Princeton, New Jersey 08540