Random Fourier Features Based Gaussian Process Models for Stochastic Simulations
Abstract
Abstract. Stochastic simulations are increasingly used to describe complex systems with uncertainties. To better characterize the uncertainty of such a simulation, we propose a random Fourier features method to fit full Bayesian Gaussian process emulators for these simulations. The random Fourier features technique uses low-dimensional features of the correlation function of the Gaussian process to achieve a low-rank approximation of the correlation matrix. We prove the convergence of the proposed random Fourier features method. Simulation results show that the proposed method can significantly reduce computation time while maintaining high prediction accuracy. The advantages of the proposed method are also illustrated using a modern subway simulation.
Fields of Study & Subjects
Gaussian Processes and Bayesian Inference
Advanced Multi-Objective Optimization Algorithms
Probabilistic and Robust Engineering Design
Publication details
- Journal / source
- SIAM/ASA Journal on Uncertainty Quantification
- Volume
- 14
- Issue
- 1
- Pages
- 48–76
- Publication date
- 6 January 2026
- Language
- EN