I am reading RBMs from this paper. In Fig1 they show an example of generating hand-written digit using RBMs. This is the figure they are showing: enter image description here

In the learning step first we sample $h$ from $h \sim P(h|v)$ and then reconstructe $v$ from $v \sim P(v|h)$. Now in the generating step they are sampling $v$ from $v \sim P(v|h)$. My question is, in generating step if we do not sample $h$ from $h \sim P(h|v)$ how can we get $P(v|h)$?



You must log in to answer this question.

Browse other questions tagged .