Abstract
Many models have addressed the development of the horizontal organization of primary visual cortex (V1), e.g. orientation and ocular dominance maps. However, the origin of vertical organization has not been addressed. Here we present a model addressing this question: why are some properties, such as preferred orientation and ocular dominance, locally invariant, while others, such as preferred spatial phase, are not? We model the development of layer 4, the input-recipient layer, of cat V1. We show that simultaneous development of geniculocortical and intracortical connections via Hebb-like synaptic plasticity, operating on a local group of mutually interconnected excitatory and inhibitory cells, along with appropriate correlations in the activities of lateral geniculate (LGN) cells, leads to (1) Development of geniculocortical connections that yield simple cells with a common preferred orientation but varying preferred spatial phases; (2) Development of intracortical connections that yield antiphase (push-pull) inhibition, and excitation among cells of similar preferred phase. The resulting circuit is one we have shown previously (Troyer et al., J Neurosci 18:5908 (1998)) to explain multiple aspects of orientation tuning, including its contrast invariance. We can generalize to the following principles: Hebb-like plasticity will guide a local piece of layer 4 to contain cells responsive to anticorrelated stimulus pairs, which mutually inhibit one another; therefore the properties that are locally invariant are those shared by anticorrelated pairs, while those that differ locally are those that differ between such pairs. Anticorrelated pairs share a common orientation and ocular dominance, but have opposite spatial phase preference, explaining the local invariances of cat V1. Contrast-invariant tuning — and more generally, magnitude-invariant form recognition — is achieved by this push-pull inhibition, provided that cortical inhibition dominates excitation.