PATTERN CLASSIFICATION USING LINEAR PROGRAMMING

Asim Roy (Inventor)

Research output: Patent

Abstract

The need for pattern classification by machine or device exists in a number of technological areas such as image classification, target recognition & tracking, speech & character recognition, signal processing, robotics, electronic & radar surveillance, medical, scientific and engineering diagnosis and like applications.Thus far, the principal approach to solving pattern classification problems involves the use of a neural network which must include two phases, a learning or training phase and a classification phase. The basic weakness of a neural network device is the fact that a long time is required for the training phase and that it involves extensive trial & error.Researchers at Arizona State University have developed a new method for pattern classification using linear programming. Although full detail on this invention can be found in US Patent No. 5,299,284, the system is different from neural networks and statistical classifiers.The training module stores a set of training examples and has a set of procedures that operate on the examples. It can be used for supervised learning where classification of the training samples are known. The output of the algorithm can be setup on a neural net-like network structure for exploiting parallelism in the classification phase. The method is robust and preliminary tests show that it is much faster than neural networks in the learning phase.
Original languageEnglish (US)
StatePublished - Jan 1 1900

Fingerprint

Linear programming
Pattern recognition
Neural networks
Surveillance radar
Character recognition
Image classification
Supervised learning
Patents and inventions
Signal processing
Robotics
Classifiers

Cite this

@misc{de8d1d7415234fe89e02c12c51b38fe1,
title = "PATTERN CLASSIFICATION USING LINEAR PROGRAMMING",
abstract = "The need for pattern classification by machine or device exists in a number of technological areas such as image classification, target recognition & tracking, speech & character recognition, signal processing, robotics, electronic & radar surveillance, medical, scientific and engineering diagnosis and like applications.Thus far, the principal approach to solving pattern classification problems involves the use of a neural network which must include two phases, a learning or training phase and a classification phase. The basic weakness of a neural network device is the fact that a long time is required for the training phase and that it involves extensive trial & error.Researchers at Arizona State University have developed a new method for pattern classification using linear programming. Although full detail on this invention can be found in US Patent No. 5,299,284, the system is different from neural networks and statistical classifiers.The training module stores a set of training examples and has a set of procedures that operate on the examples. It can be used for supervised learning where classification of the training samples are known. The output of the algorithm can be setup on a neural net-like network structure for exploiting parallelism in the classification phase. The method is robust and preliminary tests show that it is much faster than neural networks in the learning phase.",
author = "Asim Roy",
year = "1900",
month = "1",
day = "1",
language = "English (US)",
type = "Patent",

}

TY - PAT

T1 - PATTERN CLASSIFICATION USING LINEAR PROGRAMMING

AU - Roy, Asim

PY - 1900/1/1

Y1 - 1900/1/1

N2 - The need for pattern classification by machine or device exists in a number of technological areas such as image classification, target recognition & tracking, speech & character recognition, signal processing, robotics, electronic & radar surveillance, medical, scientific and engineering diagnosis and like applications.Thus far, the principal approach to solving pattern classification problems involves the use of a neural network which must include two phases, a learning or training phase and a classification phase. The basic weakness of a neural network device is the fact that a long time is required for the training phase and that it involves extensive trial & error.Researchers at Arizona State University have developed a new method for pattern classification using linear programming. Although full detail on this invention can be found in US Patent No. 5,299,284, the system is different from neural networks and statistical classifiers.The training module stores a set of training examples and has a set of procedures that operate on the examples. It can be used for supervised learning where classification of the training samples are known. The output of the algorithm can be setup on a neural net-like network structure for exploiting parallelism in the classification phase. The method is robust and preliminary tests show that it is much faster than neural networks in the learning phase.

AB - The need for pattern classification by machine or device exists in a number of technological areas such as image classification, target recognition & tracking, speech & character recognition, signal processing, robotics, electronic & radar surveillance, medical, scientific and engineering diagnosis and like applications.Thus far, the principal approach to solving pattern classification problems involves the use of a neural network which must include two phases, a learning or training phase and a classification phase. The basic weakness of a neural network device is the fact that a long time is required for the training phase and that it involves extensive trial & error.Researchers at Arizona State University have developed a new method for pattern classification using linear programming. Although full detail on this invention can be found in US Patent No. 5,299,284, the system is different from neural networks and statistical classifiers.The training module stores a set of training examples and has a set of procedures that operate on the examples. It can be used for supervised learning where classification of the training samples are known. The output of the algorithm can be setup on a neural net-like network structure for exploiting parallelism in the classification phase. The method is robust and preliminary tests show that it is much faster than neural networks in the learning phase.

M3 - Patent

ER -