Large-Scale Brain Area Functional Organization - Computational Explorations in Cognitive Neuroscience - page 202

We begin with a brief summary of some of the gen-
eral functional and computational principles that un-
derlie all of our cognitive models. Then we provide a
brief overview of the different anatomical areas of the
cortex and relevant aspects of the subcortical anatomy.
We then describe a tripartite functional organization of
these areas in terms of the following specialized sys-
tems: the posterior cortex ,the frontal cortex ,andthe
hippocampus and related structures . This tripartite
organization constitutes a kind of cognitive architec-
ture — a higher level description of the processes un-
derlying cognitive function, where different cognitive
phenomena can be explained in terms of the interac-
tions between these specialized systems (in addition to
the common principles and mechanisms applicable to
all areas).
A central organizing principle in our framework is
the notion of a tradeoff between different functional
objectives. A tradeoff is where two objectives are mu-
tually incompatible, and thus cannot be simultaneously
achieved — achieving one objective trades off against
achieving the other. Where such tradeoffs can be iden-
tified, and different brain areas associated with differ-
ent functions, we can provide a principled account for
the observed functional specialization — the brain ar-
eas are specialized to optimize separately two different
functional objectives that would otherwise conflict if a
unitary system were to try to achieve both of them. Be-
cause these tradeoffs are based on the kinds of mecha-
nisms and principles developed in the first part of this
topic, they provide a means of leveraging basic prop-
erties of neural computation to explain aspects of the
large-scale organization of the brain.
The first tradeoff we explore is in the rate of learn-
ing and the nature of the resulting representations, with
cortex (posterior and frontal) being slow and integra-
tive (integrating over instances), and the hippocampus
being fast and separating (keeping instances separate).
The second tradeoff is in the ability to update rapidly
and maintain robustly representations in an active state
over delays and in the face of interference from ongo-
ing processing ( active maintenance ). The frontal cor-
tex (and particularly the prefrontal cortex ) appears to
be specialized for this kind active maintenance, which
plays an important role in both active memory and con-

trolled processing (i.e., “executive” control of cogni-
tion).
The last part of this chapter addresses a number of
general problems that arise in using our framework to
model cognitive phenomena. This framework explains
a number of aspects of cognition quite naturally, but it
is not immediately clear how some other aspects can be
explained. We highlight these difficulties as important
issues to be addressed by the models presented in sub-
sequent chapters and in future research.

7.2 General Computational and Functional

Principles

We can usefully divide our discussion of the general
properties of cognition into their structural and dy-
namic aspects. The structural aspects describe the ways
that information and processing are arranged within the
system, determined by the overall patterns of connectiv-
ity and relationships between representations at differ-
ent levels or stages of processing. The dynamic aspects
describe the nature of processing over time, determined
by how activation flows through the various processing
levels and achieves a useful overall outcome.

7.2.1 Structural Principles

Many aspects of processing in the cortex are arranged
in a generally hierarchical fashion (i.e., having ordered
subordinate and superordinate levels), with many differ-
ent specialized pathways of such hierarchies that each
operate on and emphasize different aspects of the over-
all sensory input, motor output, or intermediate process-
ing. However, rich interconnectivity between these dif-
ferent pathways at different levels provides a number of
functional benefits, so we do not think of them as com-
pletely distinct from one another (i.e., parallel, modular,
and strictly hierarchical pathways), but rather as highly
interconnected, interdependent, and only approximately
hierarchical. It is also clear that processing and mem-
ory are embedded in the same underlying neural hard-
ware, and distributed over a potentially wide range of
different processing pathways. These properties have
a number of important consequences, as elaborated in
subsequent sections.