Finding the Maximizers of the Information Divergence from an Exponential Family
Contact the author: Please use for correspondence this email.
Submission date: 23. Dec. 2009
published in: IEEE transactions on information theory, 57 (2011) 6, p. 3236-3247
DOI number (of the published article): 10.1109/TIT.2011.2136230
MSC-Numbers: 94A17, 62B10, 52C40, 13P25
Keywords and phrases: kullback-leibler divergence, relative entropy, exponential family, information projection, optimization, commutative algebra
Download full preprint: PDF (280 kB)
This paper investigates maximizers of the information divergence from an exponential family . It is shown that the rI-projection of a maximizer P to is a convex combination of P and a probability measure with disjoint support and the same value of the sufficient statistics A. This observation can be used to transform the original problem of maximizing over the set of all probability measures into the maximization of a function over a convex subset of . The global maximizers of both problems correspond to each other. Furthermore, finding all local maximizers of yields all local maximizers of .
This paper also proposes two algorithms to find the maximizers of and applies them to two examples, where the maximizers of were not known before.