| |
Abstract:
Previous studies of category learning have shown that
subjects can learn random dot prototypes without explicit
awareness; even individuals with severely impaired explicit memory
skills performed normally. Implicit learning of non-prototype
categories however is less well studied. It is uncertain whether
these types of categories can be learned without awareness in
normal subjects, and how varying the complexity of the categories
impacts the ease of learning. We present recent results from a
normal behavioral study of 24 subjects. Four progressively more
complex categorization rules were tested: two visual patterns and
two linearly inseparable rules. There were significant learning
effects for all four rules. However, the learning effect was
smaller for the linearly inseparable rules. Furthermore, analysis
of the components of learning showed that for each of the linearly
inseparable rules, subjects learned only a linearly separable
subcomponent. This suggests that implicit learning may be limited
to learning only linearly separable functions. It is well known
that linearly inseparable functions require a hidden-layer for
Parallel Distributed Processing (PDP) learning. A PDP model was
used to illustrate how this applies to the category rules used in
these behavioral experiments. The results support a view that the
hippocampal region elaborates the neocortical representation, and
thus computationally provides a hidden layer for learning, which is
absent in implicit learning. Supported by ADRC Grant
#P50-AG05133.
|
