MAP is the optimization of prior times likelihood in bayesion method.
MAP is not invariant to paramerterization. It is a point estimation and tend to squeeze the value around the point, flat all other point.
原文：http://www.cnblogs.com/wintor12/p/4125348.html