Adaptive Surrogate Modeling for High-Dimensional Spatio-Temporal Outputs
A newly developed adaptive surrogate modeling method addresses the high computational cost of analyzing spatio-temporal multi-physics systems with very high-dimensional outputs. The technique first applies a dimension reduction method to project the high-dimensional output into a low-dimensional latent space, then constructs the surrogate model within that reduced space. Prediction error in the original space, which combines reconstruction error from the dimension reduction and error from the surrogate model itself, is evaluated using several error metrics. This approach is designed to improve computational efficiency in workflows such as uncertainty quantification and optimization, which demand numerous model evaluations. The paper is available as an arXiv preprint under identifier 2608.17250.
Key facts
- The paper develops an adaptive surrogate modeling method for high-dimensional spatio-temporal outputs.
- Spatio-temporal multi-physics systems are computationally expensive and involve many inputs and outputs.
- Surrogate models replace physics-based models to achieve computational efficiency.
- A dimension reduction method maps high-dimensional output to a low-dimensional latent space.
- The surrogate model is constructed in the low-dimensional space.
- Prediction error in the original space includes both reconstruction error and surrogate model error.
- Different error metrics are used to evaluate the prediction error.
- The method targets applications like uncertainty quantification and optimization that require many function calls.
Entities
—