Abstract
This paper presents an approach for constrained Gaussian Process (GP)regression where we assume that a set of linear transformations of the processare bounded. It is motivated by machine learning applications forhigh-consequence engineering systems, where this kind of information is oftenmade available from phenomenological knowledge, and the resulting constraintsmay be essential to achieve the level of confidence needed. We consider a GP$f$ over functions on $\mathcal{X} \subset \mathbb{R}^{n}$ taking values in$\mathbb{R}$, where the process $\mathcal{L}f$ is still Gaussian when$\mathcal{L}$ is a linear operator. Our goal is to model $f$ under theconstraint that realizations of $\mathcal{L}f$ are confined to a convex set offunctions. In particular we require that $a \leq \mathcal{L}f \leq b$ given twofunctions $a$ and $b$ where $a < b$ pointwise. This formulation provides aconsistent way of encoding multiple linear constraints, such asshape-constraints based on e.g. boundedness, monotonicity or convexity as arelevant example. We adopt the approach of using a sufficiently dense set ofvirtual observation locations where the constraint is required to hold, andderive the exact posterior for a conjugate likelihood. The results needed forstable numerical implementation are derived, together with an efficientsampling scheme for estimating the posterior process which is exact in thelimit. A few numerical examples focusing on noiseless observations are given.This is relevant for computer code emulation and is also more computationallydemanding than the alternative scenario with i.i.d. Gaussian noise.