Near-Maximum Entropy Models for Binary Neural Representations of Natural Images

Matthias Bethge & Philipp Berens
Computational Vision and Neuroscience Spotlight ID W46

What is the influence of pairwise correlations on large networks of neurons? Near-MaxEnt Model Dichotomized Gaussian distribution: Parameter estimation and sampling are easy.

Low dimensional observations: often well described with pairwise statistics Full system: ?

Binarized Natural Images
Small network: Large network: Pairwise correlations Pairwise correlations fail sufficient

HIsing ­ HDG

Model has entropy close to the maximum.