Maximum a priori estimation (MAP) is the problem of selecting the value of a latent variable $y$ that maximizes the posterior probability given observations $x$:

$$ y_\text{MAP}=\argmax_y p(y\mid x) $$

MAP vs MLE

MAP is MLE with regularization induced by the prior: if the prior is uniform ($p(\theta) = \text{const})$ then MAP reduces to MLE.

Decision-theoretic view

MAP corresponds to minimizing 0-1 loss:

$$ \ell(y,\hat y)= \begin{cases} 0 & y=\hat y\\ 1 & y\ne\hat y \end{cases} $$

Under this loss, the Bayes optimal decision rule is to choose the posterior mode. If the loss is different (e.g., Hamming loss), MAP is not necessarily optimal.

References

Likelihood function

Maximum likelihood estimation (MLE)