It seems to be a pearl of conventional wisdom that parameter learning in deep
sum-product networks is surprisingly fast compared to shallow mixture models.
This paper examines the effects of overparameterization in sum-product networks
on the speed of parameter optimisation. Using theo