A probabilistic Taylor expansion with Gaussian processes

Maximum likelihood estimation in Gaussian process regression is ill-posed