Towards Data-Efficient Cross-Device Generalization of Grad-Shafranov Equilibria via Transfer Learning Neural Operator
Researchers have developed a novel neural operator framework to accelerate the real-time reconstruction of magnetohydrodynamic equilibria in fusion devices. This approach recasts equilibrium reconstruction as a cross-device operator learning problem, mapping geometry and profile parameters directly to the poloidal flux field. The Wavelet Neural Operator architecture demonstrated strong cross-geometry performance, achieving low relative L2 errors with limited labeled data and enabling millisecond-scale inference. AI
IMPACT Enables faster, more generalizable AI models for complex scientific simulations, potentially accelerating fusion energy research.