Unified developmental model of maps, complex cells and surround modulation in the primary visual cortex
Abstract
For human and animal vision, the perception of local visual features can depend on
the spatial arrangement of the surrounding visual stimuli. In the earliest stages of visual
processing this phenomenon is called surround modulation, where the response of
visually selective neurons is influenced by the response of neighboring neurons. Surround
modulation has been implicated in numerous important perceptual phenomena,
such as contour integration and figure-ground segregation. In cats, one of the major
potential neural substrates for surround modulation are lateral connections between
cortical neurons in layer 2/3, which typically contains ”complex” cells that appear to
combine responses from ”simple” cells in layer 4C. Interestingly, these lateral connections
have also been implicated in the development of functional maps in primary
visual cortex, such as smooth, well-organized maps for the preference of oriented lines.
Together, this evidence suggests a common underlying substrate the lateral interactions
in layer 2/3—as the driving force behind development of orientation maps for
both simple and complex cells, and at the same time expression of surround modulation
in adult animals. However, previously these phenomena have been studied
largely in isolation, and we are not aware of a computational model that can account
for all of them simultaneously and show how they are related. In this thesis we resolve
this problem by building a single, unified computational model that can explain the
development of orientation maps, the development of simple and complex cells, and
surround modulation.
First we build a simple, single-layer model of orientation map development based
on ALISSOM, which has more realistic single cell properties (such as contrast gain
control and contrast invariant orientation tuning) than its predecessor. Then we extend
this model by adding layer 2/3, and show how the model can explain development of
orientation maps of both simple and complex cells. As the last step towards a developmental
model of surround modulation, we replace Mexican-hat-like lateral connectivity
in layer 2/3 of the model with a more realistic configuration based on long-range
excitation and short-range inhibitory cells, extending a simpler model by Judith Law.
The resulting unified model of V1 explains how orientation maps of simple and
complex cells can develop, while individual neurons in the developed model express
realistic orientation tuning and various surround modulation properties. In doing so,
we not only offer a consistent explanation behind all these phenomena, but also create
a very rich model of V1 in which the interactions between various V1 properties can
be studied. The model allows us to formulate several novel predictions that relate the variation of single cell properties to their location in the orientation preference maps
in V1, and we show how these predictions can be tested experimentally. Overall,
this model represents a synthesis of a wide body of experimental evidence, forming a
compact hypothesis for much of the development and behavior of neurons in the visual cortex.