Machine learning for direct spectral measurement inversion

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.