Quantization of Independent Measurements and Recognition Performance,

Abstract

It is known that for the pattern classification problem where only a finite number of training samples are available, in general performance improves, reaches a maximum, and then starts deteriorating as the number of measurements is increased. However, one of the authors has shown that for independent measurements of binary quantization, the measurement complexity can be arbitrarily increased without fear of this peaking of performance. In the paper the authors consider the case of independent measurements with arbitrary quantization. (Author)

Document Details

Document Type
Technical Report
Publication Date
Mar 01, 1972
Accession Number
AD0747706

Entities

People

  • Akshat Jain
  • Balasubramanian Chandrasekaran

Organizations

  • Ohio State University

Tags

DTIC Thesaurus Topics

  • Classification
  • Measurement
  • Recognition
  • Training

Readers

  • Approximation Theory.
  • Mathematics or Statistics
  • Neural Network Machine Learning.