Overview of our comparison of the learned coupled representations of natural images.

The representations learned by HOCCA, CCA, ICA, and whitening were compared from both statistical and biological points of view using multiple criteria. Two properties of the learned representations are of particular interest The individual representations of the two data sets, which is related to spatio-chromatic processing for a given illumination condition. The coupling (correspondence) between the two representations, which is related to adaptation to changes in the illumination. The different criteria measure different aspects of these two properties.