Let's start our discussion of inference engines with the non-Markovian methods. Under some circumstances, these methods can provide a fast and accurate enough approximation to the posterior.