Pattern-generator-driven development in self-organizing models
CNS '97 Proceedings of the sixth annual conference on Computational neuroscience : trends in research, 1998: trends in research, 1998
Learning innate face preferences
Neural Computation
Hi-index | 0.01 |
Hebbian models of orientation map formation in the primary visual cortex typically represent the cortex as an array of neurons that are excited by both ON and OFF lateral geniculate nucleus neurons. However, simple cells with only thalamic excitation exhibit tuning curves that widen with stimulus contrast. This is inconsistent with the contrast-invariant width actually observed. We propose a map formation model that achieves contrast invariance through anti-phase inhibition. We describe how inhibition between columns implements the coverage constraint to ensure orientations are uniformly sampled. We find that the model exhibits more robust receptive field formation than a simpler model without anti-phase inhibition.