Lecture 15 - Advanced Topics in Bayesian Linear Regression

Objectives

• Use the evidence approximation to estimate regression parameters that are not weights, e.g., noise variance and hyper-parameters affecting the priors of the weights.

• Use automatic relevance determination to select which basis functions to keep in a generalized linear model.

• Use standardized errors and quantile-quantile plots to assess the quality of the posterior point-predictive distribution of a regression model.