BACK to VOLUME 42 NO.5

Kybernetika 42(5):517-538, 2006.

Maximizing Multi--Information

Nihat Ay and Andreas Knauf


Abstract:

Stochastic interdependence of a probability distribution on a product space is measured by its Kullback--Leibler distance from the exponential family of product distributions (called multi-information). Here we investigate low-dimensional exponential families that contain the maximizers of stochastic interdependence in their closure.


Based on a detailed description of the structure of probability distributions with globally maximal multi-information we obtain our main result: The exponential family of pure pair-interactions contains all global maximizers of the multi-information in its closure.


Keywords: multi-information; exponential family; relative entropy; pair-interaction; infomax principle; Boltzmann machine; neural networks;


AMS: 82C32; 92B20; 94A15;


download abstract.pdf


BIB TeX

@article{kyb:2006:5:517-538,

author = {Ay, Nihat and Knauf, Andreas},

title = {Maximizing Multi--Information},

journal = {Kybernetika},

volume = {42},

year = {2006},

number = {5},

pages = {517-538}

publisher = {{\'U}TIA, AV {\v C}R, Prague },

}


BACK to VOLUME 42 NO.5