deepICMGP - Deep Intrinsic Coregionalization Gaussian Processes for Multiple Outputs

Fits two-layer deep intrinsic coregionalization Gaussian process surrogate models for multiple outputs observed at shared input locations. Uses Markov chain Monte Carlo sampling for kernel parameters and latent coordinates, with plug-in coregionalization estimates. Provides burn-in removal and thinning, posterior predictive means and within-location output covariances, and sequential experimental design using an active learning criterion. The method is described by Chang and Sung (2026) <doi:10.1615/Int.J.UncertaintyQuantification.2026062570>. Adapts code from the 'deepgp' package, with underlying deep Gaussian process methodology described by Sauer, Gramacy, and Higdon (2023) <doi:10.1080/00401706.2021.2008505>.