Exploring Generative Physics Models with Scientific Priors in Inertial Confinement Fusion

There is significant interest in using modern neural networks for scientific applications due to their effectiveness in modeling highly complex, non-linear problems in a data-driven fashion. However, a common challenge is to verify the scientific plausibility or validity of outputs predicted by a neural network. This work advocates the use of known scientific constraints as a lens into evaluating, exploring, and understanding such predictions for the problem of inertial confinement fusion.

[1]  Jeffrey A. Melby,et al.  A time-dependent surrogate model for storm surge prediction based on an artificial neural network using high-fidelity synthetic hurricane modeling , 2015, Natural Hazards.

[2]  Anuj Karpatne,et al.  Physics-guided Neural Networks (PGNN): An Application in Lake Temperature Modeling , 2017, ArXiv.

[3]  Yi Yang,et al.  Nanophotonic particle simulation and inverse design using artificial neural networks , 2018, Science Advances.

[4]  Paris Perdikaris,et al.  Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , 2019, J. Comput. Phys..

[5]  Jim Gaffney,et al.  Thermodynamic modeling of uncertainties in NIF ICF implosions due to underlying microphysics models , 2014 .

[6]  Valentin Khrulkov,et al.  Geometry Score: A Method For Comparing Generative Adversarial Networks , 2018, ICML.

[7]  Paris Perdikaris,et al.  Physics-Constrained Deep Learning for High-dimensional Surrogate Modeling and Uncertainty Quantification without Labeled Data , 2019, J. Comput. Phys..

[8]  Wojciech Zaremba,et al.  Improved Techniques for Training GANs , 2016, NIPS.

[9]  Deborah Bard,et al.  CosmoGAN: creating high-fidelity weak lensing convergence maps using Generative Adversarial Networks , 2017, Computational Astrophysics and Cosmology.

[10]  J. Lindl.  Development of the indirect‐drive approach to inertial confinement fusion and the target physics basis for ignition and gain , 1995 .

[11]  Bernhard Schölkopf,et al.  Wasserstein Auto-Encoders , 2017, ICLR.

[12]  D. K. Bradley,et al.  Metrics for long wavelength asymmetries in inertial confinement fusion implosions on the National Ignition Facility , 2014 .

[13]  Michela Paganini,et al.  CaloGAN: Simulating 3D High Energy Particle Showers in Multi-Layer Electromagnetic Calorimeters with Generative Adversarial Networks , 2017, ArXiv.

[14]  Max Welling,et al.  Auto-Encoding Variational Bayes , 2013, ICLR.

[15]  Sepp Hochreiter,et al.  GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium , 2017, NIPS.