Maximum a-posteriori

cosmos 4th November 2016 at 2:43pm
Learning theory

A learning principle that can be viewed as approximating the expected value of the output from a model, using Bayesian statistics, by only considering the hypothesis with maximum a-posteriori probability (the most likely).

video

It can be seen as formally equivalent to Maximum likelihood by multiplying the likelihood by the prior (adds the log of the prior to the log likelihood).