Artificial Intelligence & Machine Learning
Applying AI and machine learning to accelerate fusion simulations, improve data-driven modeling and uncertainty quantification, develop surrogate turbulence models, and optimize plasma control.
Aligned with the DOE priority research objectives, the main research objectives include: (i) acceleration of numerical algorithms with AI/ML methods.; (ii) develop and implement AI/ML methods to maximize information extracted from measurements, model validation and uncertainty quantification; (iii) development of surrogate and reduced-order models; (iv) Develop and implement AI/ML methods for plasma control and optimization.
Specific research goals include the development of an AI/ML unified hybrid data-driven method to accelerate Monte Carlo simulations incorporating the quantification of confinement losses in bounded domains. Develop an AI/ML framework for the construction of surrogate turbulence models based on pre-trained variational autoencoders and fine-tuning strategy using the AutoencoderKL from Stable Diffusion coupled with Long Short-Term Memory networks for temporal evolution modeling. Develop custom monotonicity-promoting priors and Bayesian Gaussian Process Regression fits to enable data-driven models to remain consistent with established physical constraints while still capturing complex diagnostic behavior.
