On degeneracy and invariances of random fields paths with applications in Gaussian process modelling
Introduction
Whether for function approximation, classification, or density estimation, probabilistic models relying on random fields have been increasingly used in recent works from various research communities. Finding their applied roots in geostatistics and spatial statistics with optimal linear prediction and Kriging (Matheron, 1963, Stein, 1999), random field models for prediction have become a main stream topic in machine learning (under the Gaussian Process Regression terminology, see, e.g., Rasmussen and Williams, 2006), with a spectrum ranging from metamodeling and adaptive design approaches in science and engineering Welch et al. (1992), Jones (2001), O’Hagan (2006) to theoretical Bayesian statistics in function spaces (see Van der Vaart and Van Zanten, 2008a, Van der Vaart and Van Zanten, 2008b, Van der Vaart and van Zanten, 2011 and references therein).
Often, a Gaussian random field model is assumed for some function of interest, and so all prior assumptions on are accounted for by the corresponding mean function and covariance kernel . The choice of and should thus reflect as much as possible any prior belief the modeller wishes to incorporate in the model. Such prior belief on may of course include classical regularity properties in the first place (continuity, differentiability, Hölder regularity, etc.), but also more specific properties such as symmetries (Haasdonk and Burkhardt, 2007, Ginsbourger et al., 2012), sparse functional ANOVA decompositions (Duvenaud et al., 2011, Durrande et al., 2012, Ginsbourger et al., 2014), or degeneracy under multivariate differential operators in the case of vector-valued random fields. To take a concrete example, covariance structures characterizing divergence-free and curl-free random vector fields have been recently presented and illustrated in Scheuerer and Schlather (2012). Besides that, the idea of expressing structure with kernels has been explored in Duvenaud (2014), where a number of practical aspects regarding positive-semidefiniteness-preserving operations are addressed.
Here we shall discuss how the two first moments influence mathematical properties of associated realizations (or paths), both in a general second order set-up and in the Gaussian case. A number of well-known random field properties driven by the covariance kernel are in the mean square sense, e.g. continuity and differentiability (Cressie, 1993). However, such results generally are not informative about the pathwise behaviour of underlying random fields. On the other hand, much can be said about path regularity properties of random field paths (see, e.g., classical results in Cramér and Leadbetter, 1967, Adler, 1990), based in particular on the behaviour of the covariance kernel in the neighbourhood of the diagonal in the second order case. In the stationary case, it is then sufficient to look at the covariance function in the neighbourhood of the origin (with similar results for the variogram in the intrinsic stationary–but not necessarily second order–case). More recently, Scheuerer (2010) has taken a new look at path regularity of second-order random fields, and drew conclusions about a.s. continuous differentiability in non-Gaussian settings. Also, we refer to Scheuerer (2009) for an enlightening exposition of state-of-the-art results concerning regularity properties of random field sample paths in various frameworks.
Our focus in the present work is on pathwise mathematical properties of second order random fields and statistical applications thereof in the context of Gaussian process modelling. Motivated by several practical situations, we pay a particular attention to random fields that are supported by the null space of some linear operator , i.e. for which As we first develop in general second-order settings, an impressive diversity of path properties including invariances under group actions or sparse ANOVA decompositions of multivariate paths can be encapsulated in the framework of Eq. (1). Furthermore, in the particular case of Gaussian random fields, a more general class of path properties (notably some degeneracy properties involving differential operators) can be covered through the link between operators on the paths and operators on the reproducing kernel Hilbert space (Berlinet and Thomas-Agnan, 2004) associated with the random field, and also through an additional representation of in terms of Gaussian measures on Banach spaces.
While Section 2 is dedicated to the exposition of the main results, proofs are presented in the Appendix to ease the reading. Applications in the context of random field modelling, and especially for Gaussian process modelling, are then investigated throughout Section 3. In particular, we tackle zero-integral random processes, random fields with paths invariant under group actions, random fields with additive paths, random fields with harmonic paths, and discuss further potential applications.
In Section 4, we present two original numerical experiments where the notions of degeneracy and invariance appear very useful in Gaussian process modelling under two types of structural prior information. In the first case, the objective function possesses an unknown axis of symmetry, which is inferred by maximum likelihood, relying on a family of argumentwise invariant covariance kernels. In the second case, we obtain an improved interpolation of a solution to the heat equation thanks to a bi-harmonic kernel. The proposed model enables performing harmonic conditional simulations, which has very beneficial consequences in terms of estimation of the maximum. Section 5 is dedicated to conclusions and perspectives. The main results are finally proven in the Appendix.
Section snippets
Main results
Let be a measurable space, be a complete probability space, and be a measurable real-valued stochastic process over . Let us further assume that the paths of belong with probability 1 to some function space , where is the set of -measurable functions, and consider a linear operator . Here both and are assumed second order, in the sense that their marginals possess a variance, and we aim at giving necessary and sufficient
On processes with paths integrating to zero
Lemma 1 of Appendix ensures that the almost sure nullity of can be characterized through the nullity of and , provided that the function is -integrable. In practice, assuming that this integrability condition is fulfilled–as it is notably the case for compact and continuous and –it suffices to check that and for arbitrary . Taking for instance the settings of Durrande et al. (2013) where is a finite (positive) measure and
Estimation of a symmetry axis
We now consider the following test function over : This function is symmetric with respect to the axis defined by the equation . Given noisy evaluation results of at 20 points, and assuming a priori that is symmetric with respect to some axis (parametrized by its angle with the ordinate axis and its distance to the origin ), we aim at both reconstructing and recovering the equation of this axis by using a
Conclusions and perspectives
This article focuses on the control of pathwise invariances of square-integrable random field through their covariance structure. It is illustrated on various examples how a number of features one may wish to impose on paths such as multivariate sparsity, symmetries, or being solution to a vast class of homogeneous ordinary or partial differential equations may be cast as degeneracy or invariance properties under linear operators.
One of the main results of this work, given in Proposition 1,
Acknowledgements
The authors would like to thank Dominic Schuhmacher, Fabrice Gamboa, as well as two anonymous referees for useful comments on previous versions of this paper. They are also very grateful to the editor for constructive remarks having led to significant improvements.
References (37)
- et al.
ANOVA kernels and RKHS of zero mean functions for model-based sensitivity analysis
J. Multivariate Anal.
(2013) Bayesian analysis of computer code outputs: A tutorial
Reliab. Eng. Syst. Saf.
(2006)Regularity of the sample paths of a general second order random field
Stochastic Process. Appl.
(2010)- et al.
Disintegration of Gaussian measures and average-case optimal algorithms
J. Complexity
(2007) - et al.
Random Fields and Geometry
(2007) - et al.
Reproducing Kernel Hilbert Spaces in Probability and Statistics
(2004) - et al.
Stationary and Related Stochastic Processes: Sample Function Properties and their Applications
(1967) Karhunen-Loève expansions of mean-centered Wiener processes
Additive covariance kernels for high-dimensional Gaussian process modeling
Ann. Fac. Sci. Toulouse
Automatic model construction with Gaussian processes
Additive Gaussian processes
Multivariate Gaussian process emulators with nonseparable covariance structures
Technometrics
The Theory of Stochastic Processes I
Argumentwise invariant kernels for the approximation of invariant functions
Ann. Fac. Sci. Toulouse
Cited by (23)
Covariance models and Gaussian process regression for the wave equation. Application to related inverse problems
2023, Journal of Computational PhysicsInference and uncertainty propagation of GB structure-property models: H diffusivity in [100] tilt GBs in Ni
2021, Acta MaterialiaCitation Excerpt :As we will show below, this approach succeeds in enforcing the desired symmetries, but at the cost of introducing topological artifacts. We propose an alternative approach that circumvents this drawback by employing an unsymmetrized distance function in conjunction with the sum-over-group-orbits strategy [39–41]. Distance metrics for GBs have been an active area of research for some time and there are a variety of candidates in the literature (see e.g. [42–46] and [47] for a recent review of extant distance metrics).