Maximum a posteriori optimization of parameters and the Laplace approximation for the marginal likelihood are both basis-dependent methods. In this note I demonstrate that in the case of models parameterized by probabilities there is a better choice for the basis than the traditional choice.
Submitted to Machine Learning October 14th 1996
Accepted pending minor modifications February 23rd 1998
Revised version completed May 11th 1998
Published Volume 33, No. 1, October 1998