Date of Original Version



Technical Report

Rights Management

All Rights Reserved

Abstract or Description

Abstract: "Many recent connectionist models can be categorized as associative memories or pattern classifiers. Viewed at the right level of abstraction, the two are the same. Connectionists sometimes appear to be trying to squeeze all of cognition into the associative memory paradigm, perhaps because it's the only thing they know how to implement with gradient descent learning algorithms. But the combinatorial structure of thought and language indicates that the answer to 'How can slow components think so fast' lies beyond mere associative recall. We must search for additional cognitive primitives that can be implemented in parallel hardware. One modest successor to associative recall is considered here."