2013Unpublished venueRequires access

Compressive Estimation of a Spatial Gaussian Process

Mehrzad Malmirchegini

Open publisher page 0 citations

Abstract

In this paper, we consider estimating the spatial variations of a field that can be represented by a Gaussian process, based on a small number of observations in a sensor network. We consider cases where the underlying dynamical model is unobservable and therefore the traditional filtering approaches are not able to estimate the field's hyper parameters. More specifically, we look at the Gaussian process with Gaussian radial basis, which is a good regression model for a smooth field. We then propose an integrated log-likelihood and sparsity-based estimator. Furthermore, we apply the random projections over the field to improve the RIC properties and the overall performance for the case of narrow Kernels. We also discuss the impact of different Gaussian field parameters on the estimation. Overall, the proposed framework can be applied to compressively sample and estimate any field that can be represented by Gaussian processes.

About this research paper

What this paper is about

In this paper, we consider estimating the spatial variations of a field that can be represented by a Gaussian process, based on a small number of observations in a sensor network. We consider cases where the underlying dynamical model is unobservable and therefore the traditional filtering approaches are not able to estimate the field's hyper parameters. More specifically, we look at the Gaussian process with Gaussian radial basis, which is a good regression model for a smooth field. We then propose an integrated log-likelihood and sparsity-based estimator. Furthermore, we apply the random projections over the field to improve the RIC properties and the overall performance for the case of narrow Kernels. We also discuss the impact of different Gaussian field parameters on the estimation. Overall, the proposed framework can be applied to compressively sample and estimate any field that can be represented by Gaussian processes.

Why it matters

A significance statement is not available in the OpenAlex record.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

In this paper, we consider estimating the spatial variations of a field that can be represented by a Gaussian process, based on a small number of observations in a sensor network. We consider cases where the underlying dynamical model is unobservable and therefore the traditional filtering approaches are not able to estimate the field's hyper parameters. More specifically, we look at the Gaussian process with Gaussian radial basis, which is a good regression model for a smooth field. We then propose an integrated log-likelihood and sparsity-based estimator. Furthermore, we apply the random projections over the field to improve the RIC properties and the overall performance for the case of narrow Kernels. We also discuss the impact of different Gaussian field parameters on the estimation. Overall, the proposed framework can be applied to compressively sample and estimate any field that can be represented by Gaussian processes.

Key concepts: Gaussian random field, Gaussian process, Estimator, Gaussian, Kriging, Random field, Compressed sensing, Unobservable

Related papers

Back to paper searchBrowse research topicsOriginal source
Compressive Estimation of a Spatial Gaussian Process — Research Paper | ScholarLens