| |
Abstract:
Damage to prefrontal cortex (PFC) can lead to a variety of
dissociable cognitive deficits (e.g., perseveration,
distractibility, disinhibition, working memory impairments, etc.).
Although these deficits are quite diverse, they share a family
resemblance and are often classified together as executive deficits
or as a dysexecutive syndrome. A central challenge is to provide an
explicit model of executive control and show how it can unify these
seemingly disparate cognitive deficits. We present a simple and
computationally explicit model of executive control that provides a
unified account of a range of executive deficits. The model assumes
massive, bidirectional connectivity, continuous-valued units, and
Hebbian learning. Together these assumptions give rise to
continuous attractor nets that settle into stored patterns of
distributed activity, but that can be modulated (enhanced and
maintained, or inhibited) by external input. The PFC module of the
model represents control signals (excite/inhibit) rather than task
content and serves to modulate activity in posterior attractor
nets. We present neural network models based on this attractor
modulation theory for three quite different tasks: delayed response
(a working memory task), letter fluency (a verbal association
task), and Stroop (a selective attention task). We show that damage
to the PFC component of the models simulates the major executive
deficits of patients with prefrontal damage.
|
