August 2014
Volume 14, Issue 10
Free
Vision Sciences Society Annual Meeting Abstract  |   August 2014
A Unified Computational Model of Primary Visual Cortex: Consolidation of the Scattered Literature on Simple and Complex Cells
Author Affiliations
  • Tadamasa Sawada
    Department of Psychology, the Ohio State University
  • Alexander A. Petrov
    Department of Psychology, the Ohio State University
Journal of Vision August 2014, Vol.14, 1189. doi:10.1167/14.10.1189
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Tadamasa Sawada, Alexander A. Petrov; A Unified Computational Model of Primary Visual Cortex: Consolidation of the Scattered Literature on Simple and Complex Cells. Journal of Vision 2014;14(10):1189. doi: 10.1167/14.10.1189.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

The response properties of neurons in V1 have been explored in many empirical studies and modeled in quantitative detail. However, the data are scattered across dozens of articles and the models employ idiosyncratic parameterization schemes tailored to fit specific data sets. Here we consolidate the fragmented results of the prior studies in terms of a unified model. The model takes static grayscale images as inputs and represents them as firing-rate patterns across a population of simple- and complex-cell-like units tuned for orientations and spatial frequencies. Divisive normalization (Heeger, 1992) accounts for surround suppression, cross-orientation and cross-frequency suppression, and various contrast and pedestal effects. The model parameters are specified in terms of a calibration procedure that makes them implementation-independent and facilitates the inclusion of the model into larger integrated systems. The model is tested on stimuli from 18 representative neurophysiological experiments, including gratings, gabors, and plaids of various sizes, contrasts, orientations, and spatial frequencies. All fits use a common set of parameter values. The properties of units (or channels) in the model agree qualitatively and often quantitatively with the properties of simple and complex V1 cells in monkeys and cats. Moreover, the model gives parameter-free accounts of some phenomena that have not been modeled in detail before. Whereas basic properties such as orientation tuning are built into the model by design, complex properties emerge from the interactions of multiple mechanisms. For example, high-contrast gratings of sub-optimal orientation produce more surround suppression compared to optimal low-contrast gratings both empirically (Tailby, Solomon, Peirce, & Metha, 2007) and in the model. Such emergent properties can only be explained in an integrated model. Last but not least, the model can be used as an off-the-shelf building block of larger models, with default parameters that are consistent with neurophysiological measurements.

Meeting abstract presented at VSS 2014

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×