Calculus of Networks of Adaptive Elements

Abstract

In response to Rome Air Development Center Purchase Request number 64–707, dated February 3, 1964, this proposal outlines a program of research at developing a calculus of networks of adaptive elements. During the past three years, important advances have been made in the field of trainable pattern classifying machines. Particularly productive has been the work on those machines based on networks of adaptive elements. These machines, here are called learning machines, often consist of networks of adaptive threshold logic units(TLU’s) . The following developments are illustrative of some of the achievements of the learning machine research program at the Stanford research institute: (a) development of a low cost, high speed, electronically adjustable weighing elements, (b) design and construction of a large scale learning machine,(c) successful application of learning machine techniques to certain weather prediction problems (d) contributions to the theory of the trainability and capacity of a threshold logic unit, (e) investigations into the mathematical theory of networks of TLU’s, and (f) development of techniques for determining basic structural features in patterns.