Computing with Connections,
Abstract
There is a rapidly growing interest in problem-scale parallelism, both as a model of animal brains and as a paradigm for VLSI. Work at Rochester has concentrated on connectionist models and their application to vision. This paper lays out a framework for dealing with such problems. The framework is built around computational modules, the simplest of which are termed p-units. We develop their properties and show how they can be applied to a variety of problems. To show how the framework can be applied to computational problems in vision, two specific examples are developed in some detail. In the first, we describe how spatially distributed data can be associated with a complex concept. In the second, we discuss the shape from shading problem and show how a global parameter, such as light source position, interacts with the calculation of a spatially distributed parameter such as surface orientation. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1981
- Accession Number
- ADA102221
Entities
People
- D. H. Ballard
- J. A. Feldman
Organizations
- University of Rochester