Coordinated optimization of visual cortical maps (II) Numerical studies

TL;DR

This study uses numerical simulations to evaluate whether coordinated optimization models can replicate the irregular spatial architecture of visual cortical maps, finding that they do not fully account for biological complexity.

Contribution

It extends previous analytical models by numerically exploring their behavior at finite thresholds and with more complex conditions, assessing their biological plausibility.

Findings

Models do not produce biologically realistic irregular maps

Attractor solutions predict highly ordered layouts near symmetry breaking

Transient states and detuned periodicities do not improve biological realism

Abstract

It is an attractive hypothesis that the spatial structure of visual cortical architecture can be explained by the coordinated optimization of multiple visual cortical maps representing orientation preference (OP), ocular dominance (OD), spatial frequency, or direction preference. In part (I) of this study we defined a class of analytically tractable coordinated optimization models and solved representative examples in which a spatially complex organization of the orientation preference map is induced by inter-map interactions. We found that attractor solutions near symmetry breaking threshold predict a highly ordered map layout and require a substantial OD bias for OP pinwheel stabilization. Here we examine in numerical simulations whether such models exhibit biologically more realistic spatially irregular solutions at a finite distance from threshold and when transients towards attractor states are considered. We also examine whether model behavior qualitatively changes when the spatial periodicities of the two maps are detuned and when considering more than 2 feature dimensions. Our numerical results support the view that neither minimal energy states nor intermediate transient states of our coordinated optimization models successfully explain the spatially irregular architecture of the visual cortex. We discuss several alternative scenarios and additional factors that may improve the agreement between model solutions and biological observations.