Information Geometry and Its Applications pp 31-49 | Cite as

# Exponential Families and Mixture Families of Probability Distributions

## Abstract

The present chapter studies the geometry of the exponential family of probability distributions. It is not only a typical statistical model, including many well-known families of probability distributions such as discrete probability distributions \(S_n\), Gaussian distributions, multinomial distributions, gamma distributions, etc., but is associated with a convex function known as the cumulant generating function or free energy. The induced Bregman divergence is the KL-divergence. It defines a dually flat Riemannian structure. The derived Riemannian metric is the Fisher information matrix and the two affine coordinate systems are the natural (canonical) parameters and expectation parameters, well-known in statistics.