FINITE AUTOMATA, PATTERN RECOGNITION AND PERCEPTRONS
A finite automaton is abstractly represented as a set function from one finite set into another. Many of the problems posed for finite automata are then simply described in terms of set functions with special properties. Some elementary results on the existence and uniqueness of such "discrimination functions" are presented. As a related example a finite autoraaton is described which can recognize a large variety of geometric patterns (or characters) when displayed in a rather general way. A finite automaton which is essentially a "perceptron" is described. In order that such a device represent a discrimination function it is shown that a specific product of two set functions must also represent a discrimination function. Some rather severe necessary conditions for solving the basic discrimination problem are then derived. Some final comments on approximate discrimination and generalizations of the basic formulation are given. (auth)
- Research Organization:
- New York Univ., New York. Atomic Energy Commission Computing and Applied Mathematics Center
- NSA Number:
- NSA-14-016937
- OSTI ID:
- 4180830
- Report Number(s):
- NYO-2884
- Country of Publication:
- United States
- Language:
- English
Similar Records
Generalized information-lossless automata. I
Statistical mechanics of cellular automata