Accession Number:

ADA273638

Title:

High-Level Connectionist Models

Descriptive Note:

Final rept.

Corporate Author:

OHIO STATE UNIV RESEARCH FOUNDATION COLUMBUS

Personal Author(s):

Report Date:

1993-10-01

Pagination or Media Count:

45.0

Abstract:

Our goals this year involved learning in connectionist networks while automatically decomposing behaviors in order to support those behaviors with modular architectures. While there has been some work in this area, we desired to have the modules fully evolve in response to the demands of the task. To accomplish this, we needed a training mechanism more robust than back propagation, so we turned towards genetic algorithms GAs. These algorithms, based on principles adopted from natural selection, allow solutions to be evolved which fit the requirements of an environment. There is an extensive body of work applying GAs to evolving neural networks, but most simply use GAs to set the weights for a fixed-structure network.

Subject Categories:

  • Computer Systems

Distribution Statement:

APPROVED FOR PUBLIC RELEASE