Dr. Marcus Noack, Research Scientist, Lawrence Berkeley National Lab
Next-Generation Gaussian Processes for Function Approximation, Uncertainty Quantification, and Decision-Making
Gaussian processes (GPs) and Gaussian-related stochastic processes are powerful tools for function approximation, uncertainty quantification, global optimization, and autonomous data acquisition due to their robustness, analytical tractability, and natural inclusion of Bayesian uncertainty estimates. Even so, Gaussian processes are often criticized for poor approximation performance and neck-breaking computational costs in real-life applications. The reason for this gap, however, is not the methodology itself but rather a user-caused lack of flexibility and domain awareness of the underlying prior probability distribution.