Statistical insights from machine learning analysis can help researchers evaluate model performance and may even provide new physical understanding.
Artificial intelligence models offer significant benefits in areas where complex simulations are required. Machine learning-based models often better predict complex instabilities for some piece of computational resource. In areas such as fusion energy, it provides real-time control of actuators to achieve more stable experimental conditions. However, its inner workings can be mysterious.
Fale-Kaga et al. employed Shapley analysis to evaluate machine learning models used in fusion experiments. Their work can give fusion researchers insight into the model, evaluate the physical basis behind the model’s decisions, and even discover clues to new physical relationships.
The model studied by the authors was designed to process inputs related to elements of the plasma profile, such as rotation, temperature, and plasma density, and output the probability that a tearing mode instability will occur. The authors’ Shapley analysis approach used statistical methods to compare the effects of different input values.
“Analysis involves looking at some inputs, changing the inputs, and seeing how the outputs change,” said author Hiro Josep Farre-Caga. “So essentially, how much does this input affect the prediction of the output?”
Using this approach, the authors identified central electron temperature and rotational peaking as the two most important predictors of tearing mode instability, with a small influence of density changes.
The researchers see this approach as a way to not only learn more about the models they use, but also about the physics at play in these complex environments.
“These machine learning models have been shown to perform well in predicting tearing modes, allowing us to extract as much information as possible,” Fale-Kaga said. “We are trying to understand how to determine whether a tearing mode occurs before the physical model does.”
sauce: “Interpreting AI for Nuclear Fusion: Application to Plasma Profile Analysis for Tearing Mode Stability” by Hiro J. Fareh Kaga, Andrew Rothstein, Rohit Sonkar, Sang-Kyung Kim, Ricardo Chauchat, Minseok Kim, Keith Erickson, Jeff Schneider, and Egemen Koremen; physics of plasma (2026). This article can be accessed from: https://doi.org/10.1063/5.0311201 .
This paper is part of the papers of the 5th International Conference on Data-Driven Plasma Science. Click here for details. here .
