Abstract:Accurately obtaining physics model information is essential for comprehending the mechanisms of physical dynamics. However, the inherent complexity of these models, especially in scenarios involving turbulence and irregular terrains, often requires significant computational resources to acquire high-fidelity (HF) data. While deep learning-based approaches have demonstrated considerable potential, they generally necessitate large volumes of training data and often face challenges in efficiently integrating physical information, such as derivatives. Leveraging the concept of multi-fidelity (MF) surrogate model and Gaussian processes (GPs), we introduce a novel approach, fast multi-fidelity Gaussian processes with derivatives (FMGPD), in which the posterior mean of the low-fidelity (LF) model replaces the prior mean of the HF model, with the relationship between the two encapsulated by an unknown function. To speed up convergence during likelihood function optimization, we also initialize the kernel of the HF model using the best hyperparameters from the LF model. This model seamlessly integrates derivative information, facilitating accurate reconstruction of general physical models by synthesizing data from multiple fidelities, including derivatives. To address computational demands, we incorporate kernel approximation techniques based on structured kernel interpolation with derivatives (D-SKI). The effectiveness of our proposed method is validated through three case studies: (1) solving a partial differential equation (PDE) describing fluid or scalar transport in a 1D domain, (2) reconstructing a high-resolution local turbulence velocity field using the Johns Hopkins forced isotropic turbulence dataset, and (3) predicting the surface of a rugged mountainous area while finding the peak via Bayesian optimization (BO). Comparative analysis against four other algorithms underscores the superiority of our approach across multiple fields and highlights its significant potential for engineering applications.

Gaussian Process Emulation for High-Dimensional Coupled Systems

Residual Gaussian Process: A Tractable Nonparametric Bayesian Emulator for Multi-Fidelity Simulations

Computer emulation with non-stationary Gaussian processes

Deep Gaussian Process Emulation and Uncertainty Quantification for Large Computer Experiments

Gaussian process emulation for discontinuous response surfaces with applications for cardiac electrophysiology models

A Graphical Multi-Fidelity Gaussian Process Model, with Application to Emulation of Heavy-Ion Collisions

Emulation of Multivariate Simulators Using Thin-Plate Splines with Application to Atmospheric Dispersion

Hierarchical Shrinkage Gaussian Processes: Applications to Computer Code Emulation and Dynamical System Recovery

Reduced Dimensional Gaussian Process Emulators Of Parametrized Partial Differential Equations Based On Isomap

Solving High Frequency and Multi-Scale PDEs with Gaussian Processes

Symplectic Gaussian process regression of maps in Hamiltonian systems

Manifold Learning for the Emulation of Spatial Fields from Computational Models.

Gaussian Process Emulators for Quantifying Uncertainty Inco2spreading Predictions in Heterogeneous Media

Towards Scalable Gaussian Process Modeling

Fast multi-fidelity gaussian processes with derivatives for complex system modeling

Mixed-input Gaussian Process Emulators for Computer Experiments with a Large Number of Categorical Levels

Sequential adaptive design for emulating costly computer codes

Towards Gaussian Process for operator learning: an uncertainty aware resolution independent operator learning algorithm for computational mechanics

Towards Efficient Modeling and Inference in Multi-Dimensional Gaussian Process State-Space Models

Gaussian Process Emulation For Big Data In Data-Driven Metamaterials Design

Physics-Informed Variational State-Space Gaussian Processes