Multi-valued control problems and mixture density network

Randa Herzallah, David Lowe

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We have proposed a novel robust inversion-based neurocontroller that searches for the optimal control law by sampling from the estimated Gaussian distribution of the inverse plant model. However, for problems involving the prediction of continuous variables, a Gaussian model approximation provides only a very limited description of the properties of the inverse model. This is usually the case for problems in which the mapping to be learned is multi-valued or involves hysteritic transfer characteristics. This often arises in the solution of inverse plant models. In order to obtain a complete description of the inverse model, a more general multicomponent distributions must be modeled. In this paper we test whether our proposed sampling approach can be used when considering an arbitrary conditional probability distributions. These arbitrary distributions will be modeled by a mixture density network. Importance sampling provides a structured and principled approach to constrain the complexity of the search space for the ideal control law. The effectiveness of the importance sampling from an arbitrary conditional probability distribution will be demonstrated using a simple single input single output static nonlinear system with hysteretic characteristics in the inverse plant model.
Original languageEnglish
Title of host publicationIntelligent Control Systems and Signal Processing 2003
Subtitle of host publicationIPV-IFAC Proceedings
EditorsAntónio E. de Barros Ruano
PublisherAustralian Academic Press
Pages387-392
Number of pages6
Volume2
ISBN (Print)978-0-08044088-0
Publication statusPublished - Apr 2003
EventIFAC International Conference on Intelligent Control Systems and Signal Processing - Faro, Portugal
Duration: 1 Apr 20031 Apr 2003

Conference

ConferenceIFAC International Conference on Intelligent Control Systems and Signal Processing
Abbreviated titleICONS
CountryPortugal
CityFaro
Period1/04/031/04/03

Fingerprint

Importance sampling
Probability distributions
Sampling
Gaussian distribution
Nonlinear systems

Keywords

  • inversion-based neurocontroller
  • Gaussian distribution
  • prediction of continuous variables
  • Gaussian model approximation
  • hysteritic transfer characteristics
  • inverse plant models
  • multicomponent distribution
  • arbitrary conditional probability distributions
  • sampling
  • Neural Networks

Cite this

Herzallah, R., & Lowe, D. (2003). Multi-valued control problems and mixture density network. In A. E. de Barros Ruano (Ed.), Intelligent Control Systems and Signal Processing 2003: IPV-IFAC Proceedings (Vol. 2, pp. 387-392). Australian Academic Press.
Herzallah, Randa ; Lowe, David. / Multi-valued control problems and mixture density network. Intelligent Control Systems and Signal Processing 2003: IPV-IFAC Proceedings. editor / António E. de Barros Ruano. Vol. 2 Australian Academic Press, 2003. pp. 387-392
@inproceedings{3018c00cc5c8445da46c44e7df2b86e7,
title = "Multi-valued control problems and mixture density network",
abstract = "We have proposed a novel robust inversion-based neurocontroller that searches for the optimal control law by sampling from the estimated Gaussian distribution of the inverse plant model. However, for problems involving the prediction of continuous variables, a Gaussian model approximation provides only a very limited description of the properties of the inverse model. This is usually the case for problems in which the mapping to be learned is multi-valued or involves hysteritic transfer characteristics. This often arises in the solution of inverse plant models. In order to obtain a complete description of the inverse model, a more general multicomponent distributions must be modeled. In this paper we test whether our proposed sampling approach can be used when considering an arbitrary conditional probability distributions. These arbitrary distributions will be modeled by a mixture density network. Importance sampling provides a structured and principled approach to constrain the complexity of the search space for the ideal control law. The effectiveness of the importance sampling from an arbitrary conditional probability distribution will be demonstrated using a simple single input single output static nonlinear system with hysteretic characteristics in the inverse plant model.",
keywords = "inversion-based neurocontroller, Gaussian distribution, prediction of continuous variables, Gaussian model approximation, hysteritic transfer characteristics, inverse plant models, multicomponent distribution, arbitrary conditional probability distributions, sampling, Neural Networks",
author = "Randa Herzallah and David Lowe",
year = "2003",
month = "4",
language = "English",
isbn = "978-0-08044088-0",
volume = "2",
pages = "387--392",
editor = "{de Barros Ruano}, {Ant{\'o}nio E.}",
booktitle = "Intelligent Control Systems and Signal Processing 2003",
publisher = "Australian Academic Press",

}

Herzallah, R & Lowe, D 2003, Multi-valued control problems and mixture density network. in AE de Barros Ruano (ed.), Intelligent Control Systems and Signal Processing 2003: IPV-IFAC Proceedings. vol. 2, Australian Academic Press, pp. 387-392, IFAC International Conference on Intelligent Control Systems and Signal Processing, Faro, Portugal, 1/04/03.

Multi-valued control problems and mixture density network. / Herzallah, Randa; Lowe, David.

Intelligent Control Systems and Signal Processing 2003: IPV-IFAC Proceedings. ed. / António E. de Barros Ruano. Vol. 2 Australian Academic Press, 2003. p. 387-392.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Multi-valued control problems and mixture density network

AU - Herzallah, Randa

AU - Lowe, David

PY - 2003/4

Y1 - 2003/4

N2 - We have proposed a novel robust inversion-based neurocontroller that searches for the optimal control law by sampling from the estimated Gaussian distribution of the inverse plant model. However, for problems involving the prediction of continuous variables, a Gaussian model approximation provides only a very limited description of the properties of the inverse model. This is usually the case for problems in which the mapping to be learned is multi-valued or involves hysteritic transfer characteristics. This often arises in the solution of inverse plant models. In order to obtain a complete description of the inverse model, a more general multicomponent distributions must be modeled. In this paper we test whether our proposed sampling approach can be used when considering an arbitrary conditional probability distributions. These arbitrary distributions will be modeled by a mixture density network. Importance sampling provides a structured and principled approach to constrain the complexity of the search space for the ideal control law. The effectiveness of the importance sampling from an arbitrary conditional probability distribution will be demonstrated using a simple single input single output static nonlinear system with hysteretic characteristics in the inverse plant model.

AB - We have proposed a novel robust inversion-based neurocontroller that searches for the optimal control law by sampling from the estimated Gaussian distribution of the inverse plant model. However, for problems involving the prediction of continuous variables, a Gaussian model approximation provides only a very limited description of the properties of the inverse model. This is usually the case for problems in which the mapping to be learned is multi-valued or involves hysteritic transfer characteristics. This often arises in the solution of inverse plant models. In order to obtain a complete description of the inverse model, a more general multicomponent distributions must be modeled. In this paper we test whether our proposed sampling approach can be used when considering an arbitrary conditional probability distributions. These arbitrary distributions will be modeled by a mixture density network. Importance sampling provides a structured and principled approach to constrain the complexity of the search space for the ideal control law. The effectiveness of the importance sampling from an arbitrary conditional probability distribution will be demonstrated using a simple single input single output static nonlinear system with hysteretic characteristics in the inverse plant model.

KW - inversion-based neurocontroller

KW - Gaussian distribution

KW - prediction of continuous variables

KW - Gaussian model approximation

KW - hysteritic transfer characteristics

KW - inverse plant models

KW - multicomponent distribution

KW - arbitrary conditional probability distributions

KW - sampling

KW - Neural Networks

M3 - Conference contribution

SN - 978-0-08044088-0

VL - 2

SP - 387

EP - 392

BT - Intelligent Control Systems and Signal Processing 2003

A2 - de Barros Ruano, António E.

PB - Australian Academic Press

ER -

Herzallah R, Lowe D. Multi-valued control problems and mixture density network. In de Barros Ruano AE, editor, Intelligent Control Systems and Signal Processing 2003: IPV-IFAC Proceedings. Vol. 2. Australian Academic Press. 2003. p. 387-392