Modified noise kernels in Gaussian process modelling of energy surfaces

Abstract

We explore the use of non homogenous noise kernels in Gaussian process modelling to improve the potential energy curve models describing stochastic electronic structure data. We use the same noise kernels on energy curves describing deterministic electronic structure data by creating non-homogenous noise model. We observe, as well as agreement between the noise of the model and the stochastic data, a strong regularisation of the curves when using artificial noises on deterministic data which improves Gaussian processes in the over fitting regime.