======== Submitted to BBS, March 3, 1998 =========

Analysing the Executive Function of the Prefrontal System: 
Towards a Network Theory

Raju S. Bapi, KDB, ERATO, JST, Japan 
Guido Bugmann, University of Plymouth, England
Daniel S. Levine, University of Texas at Arlington, USA
John G. Taylor, King's College, University of London, England

SHORT ABSTRACT 
In recent years cognitive neuropsychology has been acquiring data that
reveal network of brain areas activated in a task.  In line with this,
theories and computational models of prefrontal executive function
should be pitched at this level.  However they also need to subsume
previous ideas at a global functional level and be consistent with
details of neurobiology.  Current theories and models of prefrontal
function lack such vertical integration.  An example of vertically
integrated theory and model is proposed.  At a conceptual level the
theory describes a hierarchical but distributed executive system,
comprising three overarching subfunctions.  At a network level, these
subfunctions arise as a result of interactions among loops connecting
brain areas.

LONG ABSTRACT
Experimental knowledge on the frontal system has become more detailed
in recent years. This calls for theories describing the executive
function of the frontal system as arising from interaction of brain
areas. Thus new theories need to be of network-type, with a structure
that corresponds to the network of areas revealed by
experiments. These theories need also to subsume previous more global
functional theories. This article proposes a classification of current
theories of prefrontal function and their computational model
counterparts into conceptual, functional and network type. It is found
that theories generally lack in vertical integration and that the
horizontal integration between models and theories is sometimes
obscure.
	This article contains a functional and computational analysis
of tasks used to test frontal deficits: Set-shifting tasks, delay
tasks, conditional association learning, self-ordered sequencing and
recency discrimination tasks. While a number of anatomical and
functional details are missing and preclude the design of a
representative network model, the functional analysis leads us to
propose that the executive can be seen as hierarchical but
distributed, and comprising at least three overarching subfunctions,
all necessary for setting goals and subgoals.  These three
subfunctions are (1) establishing and connecting working memory
representations; (2) forming and deciding between complex behavioral
schemata; (3) integrating information about affective values of
stimuli and potential actions to guide responses.
	A prototype neural network theory is proposed to illustrate
such a three-part integration scheme.. This theory is based on
interactions among loops implemented in a neural network called ACTION
that connects different areas of frontal cortex with corresponding
areas of the basal ganglia and thalamus.  An example of this theory
has been used elsewhere to model delayed response and card sorting
data..  

============================================= 
Keywords: 
Higher order processes, Rule induction, Basal Ganglia, Prefrontal
loops, Sequence learning, Higher order control, Computational
Neuroscience, Frontal cortex, Frontrostriate interactions

======================================