Multimodal Data Fusion of Non-Gaussian Spatial Fields in Sensor Networks

TL;DR

This paper introduces a novel multimodal data fusion method for non-Gaussian spatial fields in sensor networks, leveraging copula processes and rank correlation to improve prediction accuracy of physical phenomena.

Contribution

It presents a new model combining Gaussian processes and copula processes for flexible dependence modeling and a robust linear estimator using rank correlation for field reconstruction.

Findings

Outperforms Pearson correlation-based models in prediction accuracy.

Allows arbitrary marginal distributions for physical phenomena.

Provides a tractable solution for dependence modeling in non-Gaussian fields.

Abstract

We develop a robust data fusion algorithm for field reconstruction of multiple physical phenomena. The contribution of this paper is twofold: First, we demonstrate how multi-spatial fields which can have any marginal distributions and exhibit complex dependence structures can be constructed. To this end we develop a model where a latent process of these physical phenomena is modelled as Multiple Gaussian Process (MGP), and the dependence structure between these phenomena is captured through a Copula process. This model has the advantage of allowing one to choose any marginal distributions for the physical phenomenon. Second, we develop an efficient and robust linear estimation algorithm to predict the mean behaviour of the physical phenomena using rank correlation instead of the conventional linear Pearson correlation. Our approach has the advantage of avoiding the need to derive intractable predictive posterior distribution and also has a tractable solution for the rank correlation values. We show that our model outperforms the model which uses the conventional linear Pearson correlation metric in terms of the prediction mean-squared-errors (MSE). This provides the motivation for using our models for multimodal data fusion.