Data augmentation for disruption prediction via robust surrogate models
Rath, K., Rügamer, D., Bischl, B., von Toussaint, U., Rea, C., Maris, A.D., Granetz, R., et al.
Disruption prediction is a badly imbalanced learning problem: disruptions are rare, expensive, and unevenly distributed across machines. This paper builds robust surrogate models of plasma discharges, uses them to generate synthetic training examples, and evaluates whether predictors trained on the augmented data hold up better.
Journal of Plasma Physics 88, 895880502 (2022).
CitationRath, K., D. Rügamer, B. Bischl, U. von Toussaint, C. Rea, A. Maris, R. Granetz, et al. "Data augmentation for disruption prediction via robust surrogate models." Journal of Plasma Physics 88, no. 5 (2022): 895880502.