A comparative evaluation of stochastic-based inference methods for Gaussian process models

Scope of this work Gaussian Process models (GPMs) are extensively used in data analysis given their flexible mod-eling capabilities and interpretability. The fully Bayesian treatment of GP models is analyti-cally intractable, and therefore it is necessary to resort to approximations. This work focuses on Markov chain Monte Carlo (MCMC) inference techniques. The hierarchical structure of GPMs and the large dimensionality of parameter and latent variable spaces pose serious challenges to the development of efficient MCMC methods for GPMs. The work employs strategies based on efficient parameterizations and efficient proposal mechanisms and compares them on simulated and real data on the basis of convergence speed, sampling efficiency, and computational cost. Gaussian Process Models- GPMs I GPMs: par latent data p(par) p(latent|par) = GP(µ(par),K(par)) p(data|latent) I GPM- Probit regression example