Locally induced Gaussian processes for large-scale simulation experiments

Gaussian processes (GPs) serve as flexible surrogates for complex surfaces, but buckle under the cubic cost of matrix decompositions with big training data sizes. Geospatial and machine learning communities suggest pseudo-inputs, or inducing points, one strategy to obtain an approximation easing that computational burden. However, we show how placement points their multitude can be thwarted by pathologies, especially in large-scale dynamic response surface modeling tasks. As remedy, porting point idea, which is usually applied globally, over a more local context where selection both easier faster. In this way, our proposed methodology hybridizes global subset-based GP approximation. A cascade strategies planning provided, comparisons are drawn related emphasis on computer surrogate applications. We extend subset component parts accuracy–computational efficiency frontier. Illustrative examples provided benchmark real-simulation satellite drag interpolation problem.