Title :
Pattern Recognition by Convolution Polynomial
Author :
Remler, Michael P.
Author_Institution :
Departments of Medicine (Neurology) and Anatomy, University of North Carolina
fDate :
5/1/1974 12:00:00 AM
Abstract :
Pattern recognition is considered as a mapping from the set of all inputs to the numbers 0 to 1. The inputs are treated as vectors. A topological group algebra over the vector space is described. The input is treated as avariable in a polynomial of that group algebra. A correspondence between inputs and numbers is established. This correspondence is used to prove that the polynomials in the algebra can represent a solution to any pattern recognition problem. When the coefficients of the polynomial are suitably chosen vectors, the natural topology of the input vector space is preserved. The importance of this approach as a basis for a completely general efficient parallel process, and practically realizable pattern recognizing machine is presented. The concept may be realized by a modular parallel process type of machinery.
Keywords :
Group algebra, neural networks, parallel process machines, pattern recognition, polynomials.; Algebra; Automata; Biological neural networks; Convolution; Feature extraction; Machinery; Network topology; Pattern recognition; Polynomials; Sensor arrays; Group algebra, neural networks, parallel process machines, pattern recognition, polynomials.;
Journal_Title :
Computers, IEEE Transactions on
DOI :
10.1109/T-C.1974.223975