Skip to content

Maximum a Posteriori Estimation

You can think it as an advanced version of maximum likelihood estimation (MLE). In MLE, we assume the parameter \boldsymbol\theta is a constant value. However, in maximum a posteriori (MAP) estimation, we assume the parameter \boldsymbol\theta is also a random variable.

Instead of maximizing p(\mathbf X\mid \boldsymbol\theta ), we would like to maximize the posterior p(\boldsymbol\theta\mid \mathbf X ). The meaning of the posterior here is that given the samples \mathbf X, the probability for the parameters to be \boldsymbol\theta.

{\hat {\boldsymbol\theta }}={\underset {\boldsymbol\theta \in \Theta }{\operatorname {arg\;max} }}\ p(\boldsymbol\theta\mid \mathbf X ) = ={\underset {\boldsymbol\theta \in \Theta }{\operatorname {arg\,max} }}\ {\frac {p(x\mid \theta )\,p(\theta )}{ \int _ {\Theta }p(x\mid \vartheta )\,p(\vartheta )\,d\vartheta }}={\underset {\boldsymbol\theta \in \Theta}{\operatorname {arg\,max} }}\ p(x\mid \theta )\,p(\theta )

As you see, we explicitly modeled the prior for \boldsymbol\theta as p(\boldsymbol\theta).