Large-Scale Brain Area Functional Organization - Computational Explorations in Cognitive Neuroscience - page 210

cortex. However, we stop short of actually including
simulations of this idea, opting instead to focus on an-
other, simpler control mechanism based on the brain
area that provides the neuromodulator dopamine to the
cortex. These two mechanisms may work in tandem, or
there may be some finer division of labor between them
that we have yet to identify.
Interestingly, the dopaminergic brain area innervat-
ing the frontal cortex, the ventral tegmental area (VTA)
is adjacent to the substantia nigra, which is part of the
basal ganglia. Thus, the VTA constitutes another impor-
tant subcortical area, one that we actually do include in
some of our models. As we discussed in chapter 6, the
VTA and substantia nigra likely play a role in control-
ling learning as well, and there may be some homolo-
gies between this kind of learning in the basal ganglia
and frontal cortex. Thus, much like the hippocampus
and cingulate cortex, the basal ganglia are richly inter-
twined with cortical processing and learning, and future
modeling work will better illuminate the precise nature
of this relationship.
Even less is known about the cognitive role of the
cerebellum than that of the basal ganglia. One specula-
tion is that the cerebellum may be specialized for timing
intervals between events. This would explain its im-
portant role in fine motor control, where the timing of
muscle firing is critical. Clearly, such timing informa-
tion could also have a cognitive role. Several detailed
models of cerebellar involvement in motor control exist
(e.g., Schweighofer, Arbib, & Kawato, 1998a, 1998b;
Contreras-Vidal, Grossberg, & Bullock, 1997).

Interestingly, both of these tradeoffs arise in the context
of memory, suggesting that the functional demands of
memory play a central role in shaping the overall cog-
nitive architecture. As such, a more detailed account of
these tradeoffs is provided in chapter 9.
The three brain systems are as follows:

Posterior cortex: consisting of the occipital, tempo-
ral, and parietal areas of the cortex. These areas are
either directly responsible for analyzing sensory in-
puts and producing motor outputs, or are higher level
association areas that serve to coordinate and inte-
grate these activities. We characterize this system as
having rich overlapping distributed representations
built up slowly through learning to capture the stable,
salient aspects of the environment and to solve the
kinds of tasks the organism is typically faced with.
Frontal cortex: consisting of the frontal lobe, which is
nearly as large as the posterior cortex in humans (but
considerably smaller in “lower” animals). Our char-
acterization of the frontal cortex is based primarily on
studies of the prefrontal cortex , which are the frontal
areas anterior of the motor areas (it is not clear if
the motor cortex should be considered on functional
grounds to be part of the frontal or posterior cortex,
see chapter 11 for more discussion). The frontal cor-
tex appears to be specialized for the active mainte-
nance of information over time, which is particularly
useful for controlled processing , where responses are
mediated by various task-specific constraints, and not
simply automatic responses to incoming stimuli (as is
more characteristic of the posterior cortex).
Hippocampus and related structures: consisting of
the the hippocampus proper and other areas which
feed into it including the entorhinal cortex and the
subiculum. The hippocampus appears to play a criti-
cal role in the rapid acquisition of novel information ,
in contrast to the slow learning in both the posterior
and frontal cortex.

7.4 Tripartite Functional Organization

We now attempt to provide a more principled frame-
work for understanding the different characteristics of
some of the brain areas described above, based on two
functional tradeoffs that are optimized by three broadly
characterized systems of brain areas. One tradeoff is
in the rate of learning and the way this interacts with
knowledge representations. The other is in the abil-
ity to update rapidly and maintain actively informa-
tion over delays and in the face of interfering stimuli,
and how this interacts with the ability to use graded,
distributed representations with dense interconnectivity.

7.4.1 Slow Integrative versus Fast Separating

Learning

Our functional analysis begins by assuming that the
cortical systems (posterior and frontal) use the learn-