EXPERIMENTS WITH A PATTERN CLASSIFICATION TECHNIQUE.
Abstract
A variety of problems in different fields of engineering can be formulated as problems of pattern classification. A set of vectors, x, belonging both to class A and class B are given. It is required to find the function f(x) such that f > 0 when x epsilon A, and f > 0 when x epsilon B. An exponentially convergent and finite iterative procedure for solution of this problem for linear f is presented. The statistical and geometrical properties of the procedure as well as methods for extension to nonlinear cases are discussed. The procedure isapplied to the design of (i) switching circuits which are linearly or quadratically separable; (ii) filters for two-dimensional picture processing involving the detection of tanks in backgrounds; and (iii) optimal feedback controller for linear dynamic systems. Results of the experiments (i - iii) confirm the potential and utility of the procedure. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 21, 1965
- Accession Number
- AD0465187
Entities
People
- A. Arcese
- C. Blaydon
- R. L. Kashyap
- Y. C. Ho
Organizations
- Harvard University