Adaptive Pattern Dichotomization

• Author(s): Stanford University. Stanford Electronics Laboratories, James Shingle Koford,
• Publisher: Stanford University
• Pages: 114
• Language: en

Description:... The purpose of this research is to make an investigation of the principles governing the ability of an adaptive linear threshold element to 'dichotomize'--i.e., to classify into two categories--patterns consisting of combinations of analog or binary variables. Automatic pattern classification is becoming increasingly important in such fields as object identification from radar or sonar returns; speech recognition and translation; and weather prediction from meteorological data. The research described combines both aspects of earlier work into a unified approach to the adaption of the linear threshold element. (Author).