Presentation 2000/2/3
Identification of Automata by Recurrent Neural Network and Clustering Method
Yoshitaka KONISHI, Takahumi OOHORI, Kazuhisa WATANABE,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Recent works have shown that recurrent networks have tha ability to learn finite state automata(FSA). However, the learned FSA often becomes unstable as longer test input strings are presented to the network. If all states of the FSA are observable, the FSA can be learned simply by the standard BP network. In this paper, we propose a new clustering method to learn a FSA, which consists of the following three phases:(1)to learn the FSA by the Jordan-type recurrent neural network(JNN);(2)to extract the FSA's states by clustering internal representations of the JNN;(3)to learn again the FSA by these extracted states and the standard BP network. Simulation results for several Tomita Grammars show that stable FSAs can be learned in a ratio from 18 to 100% by our method.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) recurent neural network / finite state automaton / identification / clustering / Tomita Grammar
Paper # NC99-84
Date of Issue

Conference Information
Committee NC
Conference Date 2000/2/3(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Identification of Automata by Recurrent Neural Network and Clustering Method
Sub Title (in English)
Keyword(1) recurent neural network
Keyword(2) finite state automaton
Keyword(3) identification
Keyword(4) clustering
Keyword(5) Tomita Grammar
1st Author's Name Yoshitaka KONISHI
1st Author's Affiliation Department of Electrical Engineering, Faculty of Engineering Hokkaido Institute of Technology()
2nd Author's Name Takahumi OOHORI
2nd Author's Affiliation Department of Electrical Engineering, Faculty of Engineering Hokkaido Institute of Technology
3rd Author's Name Kazuhisa WATANABE
3rd Author's Affiliation Department of Electrical Engineering, Faculty of Engineering Hokkaido Institute of Technology
Date 2000/2/3
Paper # NC99-84
Volume (vol) vol.99
Number (no) 612
Page pp.pp.-
#Pages 7
Date of Issue