In the industry, numerical codes are well spread to model physical phenomena at stake in the products. Computer experiments make easier the control of product performance as they are less complex to set up and run than physical experiments. However computer codes still can’t be self-sufficient as they are often computationally expensive. The standard way of proceeding is to create a surrogate model in parallel. A surrogate model is a simpler model, quicker to evaluate, that approximates the output of the computer code. It helps selecting potential interesting points to simulate through the computer code, that enables to use it more efficiently. I will first present some simple and famous surrogate models. Then I will detail the Kriging model, also called Gaussian Process Regression, viewed through its probabilistic interpretation.
