Gaussian processes in CVA
There are many approaches to model nonlinear systems mathematically which all have specific advantages and disadvantages.
Among these approaches Gaussian processes are becoming more and more significant. Especially in the automotive industry Gaussian processes are increasingly used to accelerate the development of new vehicles.
A Gaussian process is a generalized multidimensional Gaussian distribution (according to Carl Friedrich Gauß) over an infinite number of random variables. Any finite subset of them is Gaussian distributed. Gaussian processes are characterized by transparancy of the mathematical modelling process because it is mainly based on linear algebra and Gaussian error calculation. This allows a prediction of the most probable values and a specification of confidence intervals for the prediction.
The structure of a Gaussian process is determined by a likelihood function, a covariance function and a mean value function.
CVA supports various approaches and improves them during the modeling process. This enables the user to choose the optimal approach for a Gaussian process within a specific problem.
Gaussian processes in CVA