Graduate Student Seminar with Will Charles on Machine Learning

QED emission and scattering processes are critical to model accurately in astrophysical plasma simulations to obtain physical results. However, many approximations are often made to scattering calculations in large simulations, in part because it is computationally expensive to sample from arbitrary multivariate probability distributions, which is a necessary step in resolving fundamentally stochastic QED interactions. The current state of the art methods for sampling from arbitrary distributions are Monte Carlo Markov Chain techniques. In this work I describe a novel machine learning architecture for the purpose of sampling from an explicitly known probability density function. The machine learning sampling method is found to significantly outperform current sampling methods with equal or better accuracy. Secondly, I describe a generative machine learning method for function space mapping, with a specific use case in speeding up simulations of modified Newtonian gravity.