Accession Number:

ADA218939

Title:

A Connectionist/Control Architecture for Working Memory and Workload: Why Working Memory Is Not 7+ /-2

Descriptive Note:

Technical rept.

Corporate Author:

CARNEGIE-MELLON UNIV PITTSBURGH PA ARTIFICIAL INTELLIGENCE AND PSYCHOLOGY PROJECT

Personal Author(s):

Report Date:

1987-09-29

Pagination or Media Count:

8.0

Abstract:

A runnable simulation architecture for working memory is described that provides an alternative to existing models of working memory. It is used to interpret a variety of phenomena, including multiple resources, workload, chunking, sequential output, skilled and episodic memories, and stages of skill acquisition. The architecture is based on a set of modules organized into regions which communicate with each other on an innerloop of processing. A new feature of this architecture is a proposed context-storage module that temporarily stores context information in fast changing connection weights. This enables the system to expand effective working memory beyond the traditional 7 or - 2 items. The context storage system is able to reload modules after short- term information decays or is displaced in addition, it provides a means of achieving stable, robust processing under conditions of high workload. Human working memory is both very limited e.g., two digits in digit cancelling and large 80 digits skilled memory, and this range is not interpretable within the modal buffer memory models. The new connectionistcontrol architecture predicts the very limited working memory effects are the result of control processing limitations of the management of message transmission on an innerloop of processing or operating in a single region. In contrast, context storage provides a very large working memory as long as there are few associations of a given region to a given context. jhd

Subject Categories:

  • Psychology
  • Cybernetics
  • Bionics

Distribution Statement:

APPROVED FOR PUBLIC RELEASE