ABSTRACT
Model selection is a key conceptual tool for performing dimension reduction and exploiting hidden structures in the data. The general idea is to compare different statistical models corresponding to different possible hidden structures and then select among them the one that is more suited for estimation. Model selection is a very powerful theory, but it suffers in many cases from a very high computational complexity that can be prohibitive. When model selection cannot be implemented due to its computational cost, it remains a good guideline for developing computationally tractable procedures, as we will see in Chapter 4.
