Pattern Recognition Properties of Multilinear Machines,
Abstract
It is well known that pattern-recognition linear machine properties of a (finite-dimensional linear dynamical system) may be identified with the input/state map (polynomials) map to (polynomials mod characteristic polynomial). In a similar way, the pattern-recognition action of a multilinear machine is describable by projections into certain complicated equivalence classes constructed on the input space. The results reported here, obtained very recently, give a firm mathematical foundation to earlier attempts by Norbert Wiener and many others to analyze nonlinear system problems using generalized Volterra kernels and Hermite expansions. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1968
- Accession Number
- AD0731304
Entities
People
- R. E. Kalman
Organizations
- Stanford University