Next: The Representation of Up: Computational Neuroscience Previous: Efficient Simplex-Like Methods

A Model of Hyperacuity-Scale Computation in Visual Cortex by Self-Excitatory Cliques of Pyramidal Cells

Authors: [tex2html_wrap4096]D. Miller, S. Zucker

Investigator username: zucker

Category: perception

Subcategory: computational neuroscience

We describe a theory of line and edge detection in layers 2/3 of primary visual cortex. This theory provides a coherent relationship for hitherto unexplained and apparently unrelated properties of layer 2/3 neurons. For example, besides their well-known properties of coarse orientation tuning, the vast majority of layer 2/3 neurons are excitatory adaptive-spiking pyramidal cells, each connecting to thousands of other cells of the same type and with similar orientation tunings, even though they may be in iso-orientation areas separated by several mm. Many if not most of these cells are also of the S type, and thus respond to moving oriented line contours with precise spatio-temporal characteristics corresponding to each of the stimulus edges. In this paper we provide a mathematical analysis of primary visual cortex which demonstrates that these properties of superficial pyramidal cells can permit the visual system to detect line and edge contours with extremely high acuity, and in the presence of large amounts of noise. Furthermore we show, based on cortical cell counts in monkeys, that the maximum orientation acuity possible corresponds to that observed in human visual hyperacuity phenomena. Our analysis also shows that these cells, although individually unreliable as processing units, can nevertheless be extremely reliable as moderate-sized groups. We base our analysis on a dynamic analog model of computation in which the primary visual cortex stores visual contour information in the form of a very large number of small tightly interconnected groups or of excitatory S-type pyramidal cells. Each of these cliques is capable, through self-excitation initiated by lower-level afferent stimulation, of bringing itself to saturated feedback firing responses within 25 msec or less, a time which is consistent with that in which visual perception can occur

Next: The Representation of Up: Computational Neuroscience Previous: Efficient Simplex-Like Methods