Abstract:Computer models are used as a way to explore complex physical systems. Stationary Gaussian process emulators, with their accompanying uncertainty quantification, are popular surrogates for computer models. However, many computer models are not well represented by stationary Gaussian processes models. Deep Gaussian processes have been shown to be capable of capturing non-stationary behaviors and abrupt regime changes in the computer model response. In this paper, we explore the properties of two deep Gaussian process formulations within the context of computer model emulation. For one of these formulations, we introduce a new parameter that controls the amount of smoothness in the deep Gaussian process layers. We adapt a stochastic variational approach to inference for this model, allowing for prior specification and posterior exploration of the smoothness of the response surface. Our approach can be applied to a large class of computer models, and scales to arbitrarily large simulation designs. The proposed methodology was motivated by the need to emulate an astrophysical model of the formation of binary black hole mergers.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to address the challenges of modeling complex physical systems in large - scale computer experiments, especially the problems encountered when using the traditional stationary Gaussian process (GP) as a surrogate for computer models. Specifically: 1. **Non - stationarity and abrupt behavior**: The response surfaces of many computer models are not stationary but exhibit non - stationary behavior and sudden changes. Traditional stationary Gaussian processes are not very good at capturing these characteristics. 2. **Processing of large - scale data sets**: For large - scale computer experiments, the amount of data is extremely large, and traditional GP methods are difficult to scale to such a large scale. Especially when the simulation design contains millions of data points, computational efficiency becomes the main bottleneck. 3. **Specific application requirements**: This research is driven by the simulation requirements of the binary black hole merger (BBH) model (COMPAS) in astrophysics. There are discontinuous regions in the input space of this model, and only about 24% of the simulation results will produce valid outputs (i.e., binary black hole mergers). Therefore, an efficient surrogate model that can handle this complexity and large - scale data is required. To solve these problems, the paper proposes a surrogate model based on the deep Gaussian process (DGP) and its uncertainty quantification method. The main innovations include: - **Introduction of a new smoothing parameter**: By introducing a new parameter α that controls the smoothness of the DGP layers, the model can better adapt to response surfaces with different smoothness. - **Variational inference method**: Adopting the stochastic variational inference (SVI) method for inference, which allows prior setting and posterior exploration of the smoothness parameter and can be applied to simulation designs of any size. - **Non - stationary covariance function**: Using a non - stationary covariance function to capture local changes and discontinuities in the response surface. Through these improvements, this method can not only handle complex non - stationary response surfaces but also scale effectively to large - scale data sets, thus providing a powerful surrogate tool for large - scale computer experiments.

Deep Gaussian Process Emulation and Uncertainty Quantification for Large Computer Experiments

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

Computer emulation with non-stationary Gaussian processes

Gaussian Process Emulation for High-Dimensional Coupled Systems

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

EzGP: Easy-to-Interpret Gaussian Process Models for Computer Experiments with Both Quantitative and Qualitative Factors

A Framework for Controlling Sources of Inaccuracy in Gaussian Process Emulation of Deterministic Computer Experiments

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

A Fast and Calibrated Computer Model Emulator: An Empirical Bayes Approach

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

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

Conglomerate Multi-fidelity Gaussian Process Modeling, with Application to Heavy-Ion Collisions

Deep Ensemble as a Gaussian Process Approximate Posterior

Deep Learning Gaussian Processes For Computer Models with Heteroskedastic and High-Dimensional Outputs

Energetic Variational Gaussian Process Regression for Computer Experiments

ANOVA Gaussian process modeling for high-dimensional stochastic computational models

Gaussian process regression + deep neural network autoencoder for probabilistic surrogate modeling in nonlinear mechanics of solids

Beyond the Mean-Field: Structured Deep Gaussian Processes Improve the Predictive Uncertainties

Uncertainty-aware and Data-efficient Cosmological Emulation using Gaussian Processes and PCA

Fully Bayesian Differential Gaussian Processes through Stochastic Differential Equations

Quantifying the Uncertainty in Model Parameters Using Gaussian Process-Based Markov Chain Monte Carlo: An Application to Cardiac Electrophysiological Models