Learning from What's Right and Learning from What's Wrong

The concept of updating (or conditioning or revising) a probability distribution is fundamental in (machine) learning and predictive coding theory. two main approaches for doing so are called Pearl's rule Jeffrey's rule. Here we make, the first time, mathematically precise what distinguishes them: increases validity (expected value) decreases (Kullback-Leibler) divergence. This forms an instance more general distinction between from what's right wrong. difference these illustrated mock cognitive scenario.