Regressão não-linear bayesiana usando os processos t-student e gaussianos

Abstract

Modelling multivariate and univariate data with non-linear structure using parametric functions pre-define, makes the inferential process complicated, as the complexity of structure increases. An analysis through the concept of modelling using Gaussian Process non-linear regression, allows to flexibilize the process of adjustment of the model, define a Gaussian Process as a prior distribution over the regression function. This type of modelling is called as nonparametric because the inferential process is made directly in function spaces and no longer in a particular parametric’s form, thus resulting in a nonparametric bayesian model, which is very flexible and has been widely used in many areas. This work focus in presentation concepts about modelling and prediction with non-linear regression using as prior distribution a Gaussian Process and Student-t Process, which show a complete bayesian approach for the model, furthermore the work has the objective make a comparison of modelling between both Process.