Going deep, reaching further: What can Boltzmann machines efficiently represent?

Tobias Reinhart, Gemma De les Coves

While universal approximation theorems theoretically underpin the representation capabilities of machine learning models, they have two shortcomings: They do not focus on efficient…