Surrogate Model
Surrogate models are computationally efficient approximations of complex simulations, primarily used to accelerate optimization, uncertainty quantification, and design exploration in various scientific and engineering fields. Current research emphasizes developing accurate and robust surrogate models using diverse machine learning architectures, including neural networks (e.g., convolutional, recurrent, and graph convolutional networks), Gaussian processes, and normalizing flows, often incorporating techniques like Bayesian optimization and active learning to improve efficiency. The widespread adoption of surrogate models significantly reduces computational costs associated with high-fidelity simulations, enabling faster design iterations, more comprehensive uncertainty analyses, and ultimately, more efficient scientific discovery and technological advancement.
Papers
Total Uncertainty Quantification in Inverse PDE Solutions Obtained with Reduced-Order Deep Learning Surrogate Models
Yuanzhe Wang, Alexandre M. Tartakovsky
Accelerated training of deep learning surrogate models for surface displacement and flow, with application to MCMC-based history matching of CO2 storage operations
Yifu Han, Francois P. Hamon, Louis J. Durlofsky
Advancing Generalized Transfer Attack with Initialization Derived Bilevel Optimization and Dynamic Sequence Truncation
Yaohua Liu, Jiaxin Gao, Xuan Liu, Xianghao Jiao, Xin Fan, Risheng Liu
Bayesian Mesh Optimization for Graph Neural Networks to Enhance Engineering Performance Prediction
Jangseop Park, Namwoo Kang