[Paper Review] A Bayes-Sard Cubature Method

Toni Karvonen, Chris J. Oates+1 more|arXiv (Cornell University)|Apr 9, 2018

Probabilistic and Robust Engineering Design45 references4 citations

TL;DR

This paper introduces Bayes-Sard cubature, a probabilistic numerical method that combines Bayesian cubature with classical cubature rules by embedding exact integration of user-specified basis functions within a Gaussian process model. The method achieves two orders of magnitude lower error than standard Bayesian cubature in high-dimensional financial integrals, while maintaining robustness to kernel parameter misspecification through a limiting improper prior on regression coefficients.

ABSTRACT

This paper focusses on the formulation of numerical integration as an inferential task. To date, research effort has largely focussed on the development of Bayesian cubature, whose distributional output provides uncertainty quantification for the integral. However, the point estimators associated to Bayesian cubature can be inaccurate and acutely sensitive to the prior when the domain is high-dimensional. To address these drawbacks we introduce Bayes-Sard cubature, a probabilistic framework that combines the flexibility of Bayesian cubature with the robustness of classical cubatures which are well-established. This is achieved by considering a Gaussian process model for the integrand whose mean is a parametric regression model, with an improper flat prior on each regression coefficient. The features in the regression model consist of test functions which are guaranteed to be exactly integrated, with remaining degrees of freedom afforded to the non-parametric part. The asymptotic convergence of the Bayes-Sard cubature method is established and the theoretical results are numerically verified. In particular, we report two orders of magnitude reduction in error compared to Bayesian cubature in the context of a high-dimensional financial integral.

Motivation & Objective

• To address the instability and inaccuracy of point estimators in Bayesian cubature under high-dimensional settings.
• To integrate the robustness of classical cubature rules with the uncertainty quantification of Bayesian cubature.
• To develop a coherent Gaussian process framework that ensures exact integration of pre-specified basis functions.
• To establish theoretical convergence properties and validate performance on high-dimensional integrals.
• To demonstrate that Bayes-Sard cubature can endow any cubature rule with a probabilistic output when the number of basis functions matches the number of nodes.

Proposed method

• The method models the integrand as a Gaussian process with a mean function defined by a parametric regression model over user-specified basis functions.
• It employs an improper flat prior on the regression coefficients to ensure non-informative, robust estimation.
• The basis functions are chosen such that they are guaranteed to be exactly integrated by the resulting cubature rule.
• The method leverages a kernel structure $k_{\sigma}(x,x') = k(x,x') + \sigma^2 k_\pi(x,x')$, and takes the $\sigma \to \infty$ limit to emphasize the parametric component.
• The resulting weights are derived as the solution to a constrained optimization problem minimizing worst-case error in the RKHS while enforcing exactness on the basis functions.
• The method inherits the computational efficiency of kernel-based cubature and provides a full posterior distribution over the integral value.

Experimental results

Research questions

• RQ1 Can a probabilistic integration method be constructed that combines the uncertainty quantification of Bayesian cubature with the robustness of classical cubature rules?
• RQ2 How does the inclusion of exactly integrable basis functions affect the convergence and accuracy of the resulting cubature rule?
• RQ3 What is the theoretical behavior of the method in the limit of large kernel variance, and how does it relate to classical cubature rules?
• RQ4 Can the method achieve superior accuracy and robustness in high-dimensional problems compared to standard Bayesian cubature?
• RQ5 Under what conditions does the Bayes-Sard cubature method recover or generalize existing cubature rules?

Key findings

• The Bayes-Sard cubature method reduces numerical integration error by approximately two orders of magnitude compared to standard Bayesian cubature in a high-dimensional financial integral.
• The method demonstrates robustness to misspecified kernel parameters, outperforming Bayesian cubature under parameter uncertainty.
• Theoretical analysis confirms asymptotic convergence of the method under mild conditions on the basis functions and design points.
• When the number of basis functions equals the number of cubature nodes, the method can endow any cubature rule with a probabilistic output.
• The limiting behavior of the method as $\sigma \to \infty$ corresponds to a worst-case optimal cubature rule that exactly integrates the basis functions.
• The weights of the Bayes-Sard rule are equivalent to solving a constrained optimization problem that minimizes RKHS error subject to exactness on the basis functions.