Improved warning and control of the tokamak density limit via machine learning of an analytic stability boundary
Theory and Simulation of Disruptions Workshop, Princeton, New Jersey
Machine learning is often assumed to trade interpretability for accuracy. This talk argues the opposite is available in the density-limit problem: a closed-form, dimensionless stability boundary learned from a multi-machine database matches the predictive accuracy of a neural network while remaining simple enough to measure and act on in real time.