Sparse Physics-Based Gaussian Process for Multi-output Regression using Variational Inferenc
Résumé
In this paper a sparse approximation of inference for multi-output Gaussian Process models based on Varia- tional Inference approach is presented. A sparse Gaussian Process regression for multi-output models related by known mathematical relationships expressing physical constraints is explored. In Gaussian Processes a multi-output kernel is a covariance function over correlated outputs. Using a general framework for construct- ing auto- and cross-covariance functions that are asymptotically consistent with the physical laws, non-linear physical relationships among several outputs can be imposed. One major issue with Gaussian process is effi- cient inference, when scaling upto large datasets. The main contribution of this paper is to apply variational inference on these models of large datasets. Numerical results of the proposed methodology for theoretical data and flight test data on aircraft loads model are presented.