Wake sleep algorithm에 나오는 용어라서 설명을 찾아둠.

High level explanation of variational inference

Problem: (1) Given an input x, the posterior probability distribution over outputs y is too complicated to work with. Or (2) Given a training corpus x, the posterior probability distribution over parameters y is too complicated to work with.

Solution: Approximate that complicated posterior p(y | x) with a simpler distribution q(y).