Local nonlinear filters with the second-order accuracy for neural network parameter estimation

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F49777513%3A23520%2F06%3A00000014" target="_blank" >RIV/49777513:23520/06:00000014 - isvavai.cz</a>

Alternative codes found

RIV/49777513:23520/06:00000015

Result on the web

—

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

Local nonlinear filters with the second-order accuracy for neural network parameter estimation

Original language description

This paper concentrates on parameter estimation of multi-layer perceptron network by nonlinear filtering methods. Whereas the EKF is using pproximation by the first order Taylor series expansion, the UKF estimates the parameters numerically by using a set of deterministically chosen points having higher degree of accuracy than the EKF without need to compute the derivatives of the function describing the network. Comparable accuracy as by UKF could be obtained by the truncated second-order filter (TSF).TSF has not been used yet because it is considered as more computationally demanding due to need of evaluation the matrix of second order derivatives. This paper points out, that the SKF is faster than the UKF in parameter estimation of MLP network thanks to a special form of the matrix of the second order derivatives. The evaluation of estimation quality and computational demands are demonstrated by a numerical example.

Czech name

Odhad parametrů neuronové sítě filtry druhého řádu

Czech description

Odhad parametrů perceptronové sítě lokálními nelineárním filtračními metodami je zkoumán. Je porovnávána přesnost a rychlost rozšířeného Kalmanova filtru s filtrem druhého řádu a transformačním filtrem. Je ukázáno, že transformační filtr je při odhadu parametrů perceptronové sítě výrazně rychlejší než transformační filtr.

Type

D - Article in proceedings

CEP classification

BB - Applied statistics, operational research

OECD FORD branch

—

Project

Result was created during the realization of more than one project. More information in the Projects tab.

Continuities

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Publication year

2006

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Article name in the collection

Proceedings of 7th International Carpathian control conference

ISBN

80-248-1066-2

ISSN

—

e-ISSN

—

Number of pages

4

Pages from-to

165-168

Publisher name

VŠB -Technická univerzita

Place of publication

Ostrava

Event location

Rožnov pod Radhoštěm

Event date

Jan 1, 2006

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

—