Basin Sculpting a Hybrid Recurrent Feedforward Neural Network
Abstract
The architecture of a recurrent neural-network-based content-addressable memory is detailed along with companion training algorithm. The memory is designed to store vectors composed of strings of the integers 1 through 9. The performance characteristics of the model-memory capacity and basin size-are presented.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1998
- Accession Number
- ADA336386
Entities
People
- Michael J. Brabel
Organizations
- United States Army Research Laboratory