This presumes conditional independency - violations of which are common. Instead, you get to estimate the dependency between each observation as in advanced va

This presumes conditional independency - violations of which are common. Instead, you get to estimate the dependency between each observation as in advanced variants of Bayes chain rule. Ultimately, some place of the estimator will contain an assumption giving only bounded optimality.