The Geometry of Bayesian Inference
John Shortle, Max B. Mendel · 1996
Abstract This paper gives a method for generating a geometric picture of a conditional distribution such as the posterior distribution used in Bayesian inference. From such a picture, it is possible to see simultaneously how different observations of one variable affect the distribution of another variable. The method generalizes to representing a conditional distribution in arbitrary coordinates. Differential forms and their geometry are used to redraw the pictures under a change of coordinates. The posterior distribution for the energy of a simple spring-mass system is used to illustrate the method.