GRAPHICAL DATA PROCESSING RESEARCH STUDY AND EXPERIMENTAL INVESTIGATION.
Abstract
A small experimental learning MINOS I was developed in order to evaluate the feasibility of this type of machine for categorizing graphical data from maps, charts and aerial photographs. A much larger machine, MINOS II, has recently been constructed, including an optical preprocessor which provides a good match between the raw data and thelearning machine. Three rationales governing the design of masks for the preprocessor are discussed; the preprocessor is seen as performing operations special to the problem being considered, whereas the learning machine is a general categorizer. Learning machines make use of a component not in common use, a device to store an analog number, which may be read-out nondestructively as an analog level. Several types of magnetic 'weight' have been investigated. MINOS II uses a magnetic weight with second-harmonic read-out. Competitive weights use multiaperture magnetic cores, and magnetostrictive delay lines. The learning machine contains two layers of threshold logic. The first layer contains 66 association-units wire 100 input lines; there are 6732 weights in the first layer-the extras are for threshold levels. Training is by 'majority-rule' logic, implemented automatically; input signals may be either (1,-1), (1,0), or (1,0,-1). The output code may be either 6 or 9 bits, with arbitrary selection of categories. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1963
- Accession Number
- AD0460517
Entities
People
- A. E. Brain
Organizations
- SRI International