Gaussian Process Models for Nonparametric Functional Regression with Functional Responses

TL;DR

This paper introduces a Bayesian Gaussian process model for nonparametric functional regression with functional responses, improving inference and computational efficiency over existing methods within the fRKHS framework.

Contribution

It develops a Gaussian process approach with posterior mode matching the fRKHS estimator, incorporating predictive process models for scalable inference in functional data analysis.

Findings

Bayesian model provides straightforward statistical inference.

Predictive process modifications enhance computational efficiency.

Numerical results confirm the effectiveness of the proposed methods.

Abstract

Recently nonparametric functional model with functional responses has been proposed within the functional reproducing kernel Hilbert spaces (fRKHS) framework. Motivated by its superior performance and also its limitations, we propose a Gaussian process model whose posterior mode coincide with the fRKHS estimator. The Bayesian approach has several advantages compared to its predecessor. Firstly, the multiple unknown parameters can be inferred together with the regression function in a unified framework. Secondly, as a Bayesian method, the statistical inferences are straightforward through the posterior distributions. We also use the predictive process models adapted from the spatial statistics literature to overcome the computational limitations, thus extending the applicability of this popular technique to a new problem. Modifications of predictive process models are nevertheless critical in our context to obtain valid inferences. The numerical results presented demonstrate the effectiveness of the modifications.