Using automatic differentiation and neural networks for IFE to develop novel theory, expand simulation capabilities, and analyze experimental data
An underappreciated and more general purpose outcome of the machine learning revolution is the ability to write high-level but hardware-accelerated numerical programs that support automatic differentiation. Automatic differentiation enables the calculation of fast and accurate gradients. Furthermore, developing tools that support automatic differentiation also natively support the embedding of neural networks. Neural networks and gradient-based workflows can help tackle challenging problems in theoretical, computational, and experimental plasma physics. This talk will provide an overview of applications of automatic differentiation to developing kinetic plasma physics theory, to developing multi-scale simulation techniques, and to rapid parameter estimation in Optical Thomson Scattering.