Modeling the role of the
frontal lobes in sequential performance. II. Classification of sequences.
An avalanche model of motor sequence encoding, presented
in a previous article (Bapi & Levine, 1994), is extended. The
model is designed to reproduce data showing that monkeys with frontal lobe
damage can learn an invariant sequence of movements if it is rewarded,
but cannot learn to perform any one of several variations on a sequence
if all are rewarded. In this article, we add to our previous network
for sequence encoding an extra module that classifies these temporal sequences
and groups them together based on reward. This module is based on
a simplified version of ARTMAP (Carpenter, Grossberg, & Reynolds, 1992).
The network's behavior reproduces data of Brody and Pribram (1978) showing
that frontally damaged monkeys can perform a flexible sequence task if
they limit their performance to some of the possible variations on the
sequence. Analogies are drawn between the classifier layer and the
frontal lobes, and between the avalanche module and part of the basal ganglia.