Memory - Computational Explorations in Cognitive Neuroscience - page 299

At an anatomical level, the frontal cortex seems
to have a fairly distinctive “striped” pattern of con-
nectivity (Levitt, Lewis, Yoshioka, & Lund, 1993).
These stripes define regions of interconnectivity among
groups of prefrontal neurons that appear to be relatively
isolated from other such groups. These stripes are gen-
erally consistent with the idea explored previously that
the frontal representations might achieve wide attractor
basins by being more isolated. Furthermore, the recur-
rent interconnectivity within a stripe would obviously
be important for active maintenance.
Recent electrophysiological evidence from direction-
coding neurons in delayed-response tasks further sup-
ports this notion, suggesting that the frontal cortex is
composed of small groups ( microcolumns ) of neurons
that all encode the same directional information ( iso-
coding neurons) (Rao, Williams, & Goldman-Rakic,
1999). This suggests that these neurons are tightly in-
terconnected with each other, and presumably less in-
fluenced by the surrounding microcolumns of neurons
that encode other directions. Such a pattern of activ-
ity is exactly what would be expected from the isolated
representations idea. However, it is not exactly clear at
this point how to reconcile this data with the anatomical
stripes, which appear to be at a larger scale.
Physiological evidence from recordings of neurons
in monkeys performing tasks that require information to
be maintained over time suggests that the prefrontal cor-
tex supports working memory (e.g., Fuster & Alexan-
der, 1971; Kubota & Niki, 1971). For example, in the
delayed response task, food is hidden in one of two
wells, and then covered during a delay period. Then,
after the delay (which can vary, but is typically be-
tween 2 and 20 seconds), the monkey is allowed to
choose which well to uncover. Thus, the location of
the food must be maintained over the delay. Record-
ings of frontal neurons show the persistent activation
of location-coding representations over the delay. Fur-
thermore, frontal lesions cause impairments on delayed
response tasks (Fuster, 1989; Goldman-Rakic, 1987).
Neural recording data from Miller et al. (1996) more
specifically identifies the prefrontal cortex as the lo-
cation where activation-based memories can be main-
tained in the face of possible interference from ongoing
processing (i.e., robust active maintenance).

Question 9.11 (a) Describe what happens when the
Input2 pattern is presented. (b) Now try a
wt_scale of 3 instead of 2. What happens with
Input2 ? (c) Explain why changing wt_scale has
the observed effects.

You should have observed that by changing the rela-
tive strength of the recurrent weights compared to the
input weights, you can alter the network's behavior
from rapid updating to robust maintenance. This sug-
gests that if the relative strength of these connections
could be dynamically controlled (e.g., by a specialized
controller network as a function of prior input stimuli),
then an activation-based memory system could satisfy
the unique demands of working memory (i.e., robust
maintenance and rapid updating).

Go to the PDP++Root window. To continue on to
the next simulation, close this project first by selecting
.projects/Remove/Project_0 . Or, if you wish to
stop now, quit by selecting Object/Quit .

9.5 The Prefrontal Cortex Active Memory System

The preceding explorations have established a set of
specializations that we would expect an activation-
based working memory system to have relative to our
generic model of the posterior cortex, namely: rela-
tively wide attractor basins (e.g., through the use of
more isolated self-maintaining connectivity) and dy-
namic control of rapid updating versus robust mainte-
nance. There is a growing body of evidence that sup-
ports the idea that the prefrontal cortex (and perhaps the
entire frontal cortex) does have these kinds of special-
izations. Furthermore, it has been relatively well es-
tablished using a variety of methodologies that the pre-
frontal cortex plays a central role in working memory
(e.g., Fuster, 1989; Goldman-Rakic, 1987; Miller et al.,
1996; O'Reilly et al., 1999a). Thus, there is converg-
ing evidence that the prefrontal cortex is a specialized
area for activation-based working memory. In this sec-
tion, we review some of the relevant evidence, and then
explore a simple model of the frontal activation-based
working memory system.

In these