Networks of Neurons - Computational Explorations in Cognitive Neuroscience - page 70

by different types of neurons with different patterns of
connectivity. This separation is useful for understand-
ing the basic principles governing network function.
We explore the properties of two major types of ex-
citatory connectivity, unidirectional and bidirectional .
The unidirectional or feedforward processing of infor-
mation via excitatory interactions performs informa-
tion processing transformations essential for cognition.
Bidirectional connectivity has many advantages over
unidirectional connectivity, and is predominant in the
cortex. However, it requires inhibition to control posi-
tive excitatory feedback.
The effects of cortical inhibition can be summarized
by a simplifying inhibitory function, which we then use
to explore bidirectional excitation in the context of inhi-
bition. In the end, it will become clear how both excita-
tion and inhibition, although separable, are intricately
intertwined in overall network behavior. It is possi-
ble to summarize the overall effects of these interac-
tions in terms of constraint satisfaction , where networks
achieve a state of activation that simultaneously maxi-
mizes the satisfaction of external constraints from the
environment and internal constraints from patterns of
weights connecting the neurons.

structure that applies across all of the different cortical
areas. Thus, the functional specializations of the cor-
tex seem to take place within a common network struc-
ture, which can be simulated within a common compu-
tational framework. The properties of this structure are
the topic of this section.
The detailed description of the biological properties
of cortex is the topic of entire topics (e.g., White,
1989a) — we will condense these details considerably
here. First, there are two general classes of neurons that
have been identified in the cortex: excitatory neurons
that release the excitatory neurotransmitter glutamate ,
and inhibitory neurons that release the inhibitory neu-
rotransmitter GABA (see section 2.4.4 for details on
these transmitters).
There are two primary subtypes of excitatory neu-
rons, the pyramidal and spiny stellate neurons, and a
larger number of different subtypes of inhibitory neu-
rons, with the chandelier and basket being some of the
most prevalent (figure 3.1). The excitatory neurons con-
stitute roughly 85 percent of the total number of neurons
in the cortex (White, 1989a), and are apparently re-
sponsible for carrying much of the information flow, be-
cause they form long-range projections to different ar-
eas of the cortex and to subcortical areas. Thus, most of
the following discussion of connectivity is focused on
these excitatory neurons. Although the inhibitory neu-
rons receive both long-range and localized inputs, they
project within small localized areas of cortex, which is
consistent with their hypothesized role of providing in-
hibitory regulation of the level of excitation in the cortex
(section 3.5).
Cortical neurons are organized into six distinct layers
(figure 3.2). The six cortical layers have been identi-
fied on anatomical grounds and are important for under-
standing the detailed biology of the cortex. However,
for our purposes, we can simplify the picture by consid-
ering three functional layers: the input , hidden ,and
output layers (figure 3.3). We will use the term layer
to refer to these functional layers, and the term corti-
cal layer for the biologically based layers. As with the
kinds of simplifications of the biology that we made in
the previous chapter, one should regard this simplifica-
tion into functional layers as only a rough (but useful)
approximation to the underlying biology.

3.2 General Structure of Cortical Networks

The cerebral cortex or neocortex forms the outer part
of the brain and is most enlarged in humans relative to
other mammals. An abundance of data supports the idea
that this is where much of the neural activity underlying
cognition takes place (nevertheless, it is important to re-
member that the cortex depends critically on many other
subcortical brain areas for its proper functioning). The
cortex can be divided into a number of different cortical
areas that are specialized for different kinds of process-
ing — as we will see in the second half of this topic,
some areas are critical for recognizing objects, others
process spatial information, and still others perform lan-
guage processing, higher level planning, and the like.
Despite the apparent functional specialization of the
cortex, we are able to use a common set of principles
in modeling the cognitive functions associated with a
wide range of cortical areas. This can be attributed to
the fact that the cortex has a fairly consistent general