Outer Product Calculations.
Abstract
This report is presented in four sections. In Section 1 an analysis of the effects of noise on associative memories is presented. Special emphasis is given to generalized inverse memories. In Section 2 of this report, several algorithms which are appropriate for implementation on the NRL spatial light modulator were compiled. These include nonlinear associative memory models and a pseudo inverse memory model that is optimum for incomplete input patterns. In section 3 of this report an analysis of a memory model that is optimum in the least squares sense for input patterns with missing components was analyzed. This memory was shown to be derivable from a different optimization principle. This optimization produces that memory which has a minimum average noise output for a given error in recall. In Section 4 of this report we discuss projection techniques and subspace methods. These approaches will allow the design of dynamic memory and recall schemes that are nonlinear, have local connectivity and can be robust against distorted input patterns. Partial contents: Analysis of the effects of noise on generalized inverse and correlation matrix associative memories; Adaptive associative memory models with potential optical implementation; On optimal associative recall from an incomplete input vector; and Projection operators and subspace methods of pattern recognition.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1986
- Accession Number
- ADA167413
Entities
People
- John Gruninger
Organizations
- Aerodyne Research