Finite Difference Recursive Update on Decomposed RBF Networks for System Identification with Lost Packet

Andryani, N.A.C. and Asirvadam, V.S. and Hamid, N.H. (2009) Finite Difference Recursive Update on Decomposed RBF Networks for System Identification with Lost Packet. In: Soft Computing and Pattern Recognition, 2009. SOCPAR '09. International Conference of.

[thumbnail of FiniteDiffRBF.pdf] PDF
FiniteDiffRBF.pdf
Restricted to Registered users only

Download (425kB)

Abstract

Radial basis function networks (RBF) is one form of feed forward neural network architecture which is popular besides multi layer preceptor (MLP). It is widely used in identifying a black box system. Finite difference approach is used to improve the learning performance especially in the non-linear learning parameter update for identifying system with lost packet in online manner. Since initializing of non-linear learning's parameters is crucial in RBF networks' learning, some unsupervised learning methods such as, K-means clustering and fuzzy C-means clustering are used besides random initialization. All the possible combination methods in the initialization and updating process try to improve the whole performance of the learning process in system identification with lost packet compared to extreme learning machine as the latest improved learning method in RBF network. It can be shown that finite difference approach with dynamic step size on decomposed RBF network with recursive prediction error for the non-linear parameter update with appropriate initialization method succeed to perform better performance compared to ELM.

Item Type: Conference or Workshop Item (Paper)
Uncontrolled Keywords: K-means clustering;black box system;decomposed RBF networks;extreme machine learning;feed forward neural network architecture;finite difference recursive update approach;fuzzy C-means clustering;multilayer preceptor;nonlinear learning parameter update;packet lost;radial basis function networks;recursive prediction error;system identification;unsupervised learning methods;finite difference methods;identification;multilayer perceptrons;radial basis function networks;recurrent neural nets;recursive functions;unsupervised learning;
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering
Q Science > QA Mathematics > QA76 Computer software
Departments / MOR / COE: Centre of Excellence > Center for Intelligent Signal and Imaging Research
Depositing User: Dr Vijanth Sagayan Asirvadam
Date Deposited: 22 Nov 2012 02:55
Last Modified: 19 Jan 2017 08:25
URI: http://scholars.utp.edu.my/id/eprint/4636

Actions (login required)

View Item
View Item