Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Statistisk signalklassificering utan lärare

Lindquist, Hans and Ericsson, Staffan (1972) In MSc Theses
Department of Automatic Control
Abstract
This report deals with the problem of learning to classify observations from a mixture of random variables. The distributions of the random variables are of known functional forms but with unknown parameters. <br><br> This problem is equivalent to that of determining separating hypersurfaces in the range of the random variables. Any rule of classification will be a function of random variables and therefore itself a random variable. Using a statistical approach it is possible to construct a classification rule that will converge to an optimal one with probability one. <br><br> For computational simplicity only the simplest possible case is trated in detail, namely that of mixtures of two one-dimensional distributions. The algorithm is... (More)
This report deals with the problem of learning to classify observations from a mixture of random variables. The distributions of the random variables are of known functional forms but with unknown parameters. <br><br> This problem is equivalent to that of determining separating hypersurfaces in the range of the random variables. Any rule of classification will be a function of random variables and therefore itself a random variable. Using a statistical approach it is possible to construct a classification rule that will converge to an optimal one with probability one. <br><br> For computational simplicity only the simplest possible case is trated in detail, namely that of mixtures of two one-dimensional distributions. The algorithm is tested by simulations on different mixtures of two-parametric distributions. <br><br> An algorithm is constructed that computes the theoretical lower bound of the variance of the classification rule. Numerical results are obtained for a number of cases. (Less)
Please use this url to cite or link to this publication:
author
Lindquist, Hans and Ericsson, Staffan
supervisor
organization
year
type
H3 - Professional qualifications (4 Years - )
subject
publication/series
MSc Theses
report number
TFRT-5107
ISSN
0346-5500
language
Swedish
id
8850587
date added to LUP
2016-03-29 16:02:07
date last changed
2016-03-29 16:02:07
@misc{8850587,
  abstract     = {{This report deals with the problem of learning to classify observations from a mixture of random variables. The distributions of the random variables are of known functional forms but with unknown parameters. <br><br> This problem is equivalent to that of determining separating hypersurfaces in the range of the random variables. Any rule of classification will be a function of random variables and therefore itself a random variable. Using a statistical approach it is possible to construct a classification rule that will converge to an optimal one with probability one. <br><br> For computational simplicity only the simplest possible case is trated in detail, namely that of mixtures of two one-dimensional distributions. The algorithm is tested by simulations on different mixtures of two-parametric distributions. <br><br> An algorithm is constructed that computes the theoretical lower bound of the variance of the classification rule. Numerical results are obtained for a number of cases.}},
  author       = {{Lindquist, Hans and Ericsson, Staffan}},
  issn         = {{0346-5500}},
  language     = {{swe}},
  note         = {{Student Paper}},
  series       = {{MSc Theses}},
  title        = {{Statistisk signalklassificering utan lärare}},
  year         = {{1972}},
}