Presentation 1999/10/21
Multistage Structure of Randomized ANNs and Its Reliability Prediction Model
Eiji SUZUKI, Tohru NAKAGAWA, Hajime KITAGAWA,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In case that an input pattern is unexpected one, namely, it completely differs from that of the training set, the reliability of an ANN (Artificial Neural Network) pattern classifier is very low. The reliability can be improved by a multistage structure of randomized ANNs, in which multiple ANNs are combined in serial and parallel ways. The ANNs have different initial conditions for learning each other. In this paper, we propose a novel model for predicting the reliability of the combined multiple ANN pattern classifier. We also evaluate the model by an experiment on a pattern classification of SIN waves, and find that our model predicts the reliability well especially when the ANNs are combined in parallel way.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Randomized ANN / Multistage / Reliability / Prediction Model / Pattern Classification
Paper # NC99-45
Date of Issue

Conference Information
Committee NC
Conference Date 1999/10/21(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) Multistage Structure of Randomized ANNs and Its Reliability Prediction Model
Sub Title (in English)
Keyword(1) Randomized ANN
Keyword(2) Multistage
Keyword(3) Reliability
Keyword(4) Prediction Model
Keyword(5) Pattern Classification
1st Author's Name Eiji SUZUKI
1st Author's Affiliation Toyota Technological Institute()
2nd Author's Name Tohru NAKAGAWA
2nd Author's Affiliation Toyota Technological Institute
3rd Author's Name Hajime KITAGAWA
3rd Author's Affiliation Toyota Technological Institute
Date 1999/10/21
Paper # NC99-45
Volume (vol) vol.99
Number (no) 382
Page pp.pp.-
#Pages 8
Date of Issue