Large-scale local surrogate modeling of stochastic simulation experiments

Gaussian process (GP) surrogate modeling for large computer experiments is limited by cubic runtimes, especially with data from stochastic simulations input-dependent noise. A popular workaround to reduce computational complexity involves local approximation (e.g., LAGP). However, LAGP has only been vetted in deterministic settings. recent variation utilizing inducing points (LIGP) additional sparsity improves upon on the speed-vs-accuracy frontier. The authors show that another benefit of LIGP over (local) nugget estimation responses more natural, when designs contain substantial replication as common attempting separate signal Woodbury identities, extended replicates, afford efficient computation terms unique design locations only. This increases amount (i.e., neighborhood size) may be incorporated without flops, thereby enhancing statistical efficiency. Performance authors' upgrades illustrated benchmark and real-world simulation experiments, including an options pricing control framework. Results indicate provides accurate prediction uncertainty quantification varying dimension strategies versus modern alternatives.