The mammalian visual system is very effective in detecting the orientations of lines and most neurons in primary visual cortex selectively respond to oriented lines and form orientation columns [14,15]. Why is the visual system organized as such? We believe that the visual system is self-organized, in both long term development and short term adaptation, to ensure the optimal information processing.
Linsker applied Hebbian learning to model the development of orientation selectivity and later proposed a principle of maximum information preservation in early visual pathways [16,17]. The focus of his work has been on the feedforward connections and in his model the feedback connections are isotropic and unchanged during the development of orientation columns; but the actual circuitry of visual cortex involves extensive, columnar specified feedback connections which exist even before functional columns appear in cat striate cortex [12,18].
Our earlier research emphasized the important role of the feedback connections in the development of the columnar structure in visual cortex. We developed a theoretical framework to help understand the dynamics of Hebbian learning in feedback networks and showed how the columnar structure originates from symmetry breaking in the development of the feedback connections (intracortical, or lateral connections within visual cortex) [5,6,10].
Many aspects of our theoretical predictions agree qualitatively with neurobiological observations in primary visual cortex. Another way to test the idea of optimal information processing or other self-organization theory is through quantitative psychophysical studies. The idea is to look for changes in perception following changes in input environments. The psychophysical experiments on orientation illusions offer some opportunities to test our theory on orientation selectivity [8,9].
Orientation illusions are the effects that the perceived orientations of lines are affected by the neighboring (in time or space) oriented stimuli, which have been observed in many psychophysical experiments and were attributed to the inhibitory interactions between channels tuned to different orientations [2,4,11,20]. But there is no unified and quantitative explanation. Neurophysiological evidences support our earlier computational model in which intracortical inhibition plays the role of gain-control in orientation selectivity [22]. But in order for the gain-control mechanism to be effective to signals of different statistics, the system has to develop and adapt in different environments.
In this paper we examine the implication of the hypothesis that the intracortical connections dynamically decorrelate the activities of cortical cells, i.e., the intracortical connections are actively adapted to the visual environment, such that the output activities of cortical cells are decorrelated. The dynamics which ensures such decorrelation through associative learning is outlined in the Associative Decorrelation ... section as the theoretical framework for the development and the adaptation of intracortical connections. In the Development of ... section , the numerical simulation of the development of orientation columns is briefly presented. The quantitative comparisons of the theory and the experiments are presented in the Quantitative Comparison ... section. The dynamics and the simulations also apply to the feedforward connections. But through out this paper, the emphases is given to the feedback connections.