Nearest-Neighbor Gaussian Process
The NNGP is used because full Gaussian processes have roughly cubic covariance matrix costs. The latent response function is given a nearest-neighbor Gaussian process prior. The GP variance is fixed at one so beta controls process variance…
1 sources - 5 claims
The NNGP is used because full Gaussian processes have roughly cubic covariance matrix costs. The latent response function is given a nearest-neighbor Gaussian process prior. The GP variance is fixed at one so beta controls process variance and helps address nonidentifiability. The NNGP factorizes the joint density into conditional densities over nearest-neighbor sets. The response lengthscale is estimated by posterior updating after assigning a prior.