Which classes of functions can quantum machine learning models actually learn?

A lot of work in quantum machine learning focuses on how to practically train quantum models, or how to prove that they can be classically intractable. However, an important question is which types of functions they can actually express, and what we can conclude about their generalization power. This talk will give an overview of what we know about the types of models that generic quantum circuits represent, why they are theoretically very promising, but why they are not powerful just by virtue of being "quantum".