The project's goal is to create theoretical and computational tools to explain "brain computations" taking place in the mammalian primary visual cortex, the first location along the visual pathway in which individual nerve cells (neurons) "recognize" elementary features of the visual scene, such as pattern orientation. A large-scale, biologically realistic numerical neuronal network model will be developed for simulating how the primary visual cortex acts as a "computer" to make this feature recognition possible, and used to simulate a patch of the primary visual cortex of about 5 millimeters by 5 millimeters in size and containing close to a million neurons. Coarse-grained representations of the primary visual cortex will also be developed, which treat it as a continuum rather than a set of individual neurons and incorporate the statistics of a multitude of experimental runs in the description instead of just one, thus eliminating the need for large numbers of simulations. Finally, hybrid representations will be developed, in which select groups of neurons are described by a large-scale neuronal model, while others are described by a bulk coarse-grained representation. Such representations appear particularly promising for simulating realistic neuronal processing of stimuli in yet larger portions of the brain.

The models and software will be validated and showcased on two striking examples of observed processing in the primary visual cortex: patterns of spontaneous cortical activity and motion illusions. The former were seen to encompass collective behavior of thousands of neurons on millimeter scales, and appear to get activated in areas in which neurons prefer the same orientation. In a popular motion illusion, showing a small square immediately followed by a long bar makes the square appear as if it is "growing" to become the bar. A physiological mechanism for this illusion was observed experimentally to be actual cortical activity corresponding to the perceived "growth" of the square into the bar, similar to that caused by real motion. One intended task for the software and models is to explain such mechanisms governing the dynamical behavior of the primary visual cortex and underlying these and other cortical phenomena.

How information is processed in the visual cortex is one of the most challenging questions in neuroscience. The proposed study of spatiotemporal activity over large scales, using a very large computational model of the primary visual cortex as well as coarse-grained theoretical methods, addresses an urgent need for scale-up in modern neuroscience, which is the theoretical complement to the recent development of multi-mode, large-scale experimental methods. The models and software developed in this project are aimed at obtaining qualitatively and quantitatively realistic explanations of the biological mechanisms that underlie neuronal computations in the primary visual cortex, and possibly other cortical areas. The software will be made readily accessible to a large group of researchers across the neurosciences, with the aim that its results will help science make significant inroads into theoretical understanding of the mechanisms of sensory perception and possibly other brain functions.

Agency
National Science Foundation (NSF)
Institute
Division of Mathematical Sciences (DMS)
Type
Standard Grant (Standard)
Application #
0506396
Program Officer
Thomas F. Russell
Project Start
Project End
Budget Start
2005-08-15
Budget End
2009-07-31
Support Year
Fiscal Year
2005
Total Cost
$842,807
Indirect Cost
Name
New York University
Department
Type
DUNS #
City
New York
State
NY
Country
United States
Zip Code
10012