Higher-Level Cognition - Computational Explorations in Cognitive Neuroscience - page 394

Perseverations in the Model

age, or a less perfect division of dimensional and featu-
ral representations, could obscure such an effect in the
monkeys.

16

Intact
Feat
Dim

14

To stop now, quit by selecting Object/Quit in the
PDP++Root window.

12

, !

10

8

11.4.3 Summary and Discussion

6

4

This simulation provides a first step toward charac-
terizing the kinds of control mechanisms that enable
activation-based processing to be more flexible and dy-
namic than weight-based processing. The model ex-
tends the simpler working-memory model from chap-
ter 9 by accounting for empirical data that implicates
the frontal cortex in more flexible, less perseverative
processing.
More specifically, we found that by simulating the
role of dopamine in regulating the frontal cortex in
terms of an adaptive-critic mechanism, a rapid trial-and-
error searching process emerged. This searching pro-
cess deactivated the prefrontal cortex when errors were
made, and activated it either through noise or when per-
formance was successful. The model matched at a qual-
itative level the effects of orbital and lateral prefrontal
cortex damage on the dynamic categorization task by
encoding more detailed feature-level information in or-
bital areas, while encoding more abstract dimension-
level information in lateral areas. These different levels
of representation, when combined with the trial-and-
error control mechanism, provided a quick way of re-
configuring the categorization rule used by the network
via top-down biasing (as in the Stroop model) of differ-
ent representations. This biasing was specifically ben-
eficial when the rules were reversed, in which case the
cortical system had a difficult time overcoming the prior
(dominant) pattern of responding without the help of
top-down activity.
Although the model is successful at a qualitative,
demonstration-of-principles level, its detailed patterns
of behavior do not precisely match those of the mon-
keys in the Dias et al. (1997) studies. The monkeys ex-
perienced two different training sequences across two
different experiments that involved a sequence of in-
tradimensional shifts and reversals, and an extradimen-
sional shift/reversal. To implement these sequences in

2

0

IDS IDR EDS

Figure 11.15: Perseverations from different types of sim-
ulated frontal lesions in the model. Feat is lesions of the
feature-level PFC representations, which correspond to the or-
bital lesions (ventromedial) in the Dias et al. (1997) monkeys
— intradimensional reversals (IDR) are selectively impaired.
Dim is lesions of the dimension-level PFC representations,
which correspond to the lateral lesions in monkeys — extradi-
mensional shift/reversals (EDS) are selectively impaired. In-
tradimensional shifts (IDS) are never impaired.

picture. Figure 11.15 shows a graph of the averages
for 10 runs on each task, plotted in a manner compara-
ble with figure 11.11 showing the empirical data from
Dias et al. (1997). The qualitative patterns of effects of
PFC damage are comparable in the simulation and the
monkeys. Feature (orbital) lesions selectively impair
intradimensional reversal, dimensional (lateral) lesions
selectively impair extradimensional reversal, and nei-
ther lesion affects intradimensional shift. These effects
demonstrate that the model captures the essential con-
tribution of the PFC in facilitating more flexible (less
perseverative) processing via activation-based process-
ing.
One interesting difference between the model and
monkey data is that the featural (orbital) lesions appear
to improve extradimensional shift performance (EDS).
This effect can be attributed to the fact that top-down ac-
tivation organized not along the direction of the reversal
can actually impair performance (i.e., the featural level
organization impairs reversal across dimensions), such
that a lesion of this area can actually improve perfor-
mance slightly. Although this effect makes sense in the
model, it is likely that other collateral effects of dam-