Survey of Neural Net Paradigms for Specification of Discrete Networks.
Abstract
Innovative neural network architectures are seen as promising breakthroughs in key problems of interest in image and speech recognition, knowledge base coding and pattern classification. Cost characteristics dictate further research into massively parallel architectures. A survey of some salient characteristics of various paradigms is undertaken in the hope of extracting key underlying organizing principles. There are critiques of continuous-type systems, comments on noise, and some cognitive perspectives for discrete networks. There is a brief discussion of memory function. Some stochastic and functional outlines are given. Keywords: Cognition; Stochastic processes; Neural networks; Information theory; Discrete networks.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 31, 1988
- Accession Number
- ADA192682
Entities
People
- Michael Dvorak