ABSTRACT
In Chapter 4 we have seen how data are used to transform a “specification” into a “description”: the free parameters of the distributions are instantiated with the data, making the joint distribution computable for any value of the variables. This identification process may be considered as a learning mechanism allowing the data to shape the description before any inferences could be made. In this chapter, we consider learning problems in more detail and show how some of them may be expressed as special instances of Bayesian programs.
