Presentation 1996/3/18
On Bayes Optimal Prediction based on Mixture Model of Multilayer Neural Networks
Hiroki Hashikawa, Masayuki Gotoh, Nobuhiko Tawara,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) In learning of the probabilistic models, it is important to predict accurately for the output of future observation. On prediction of the future observation, it is not necessary to select a particular model from the model class. The objective here is to predict the output of the future observations accurately. On the other hand, a lot of researches of the prediction methods based on Bayes decision theory for the probabilistic models have been reported. These Bayesian methods are efficient to the prediction with accuracy. In this paper, we, at first, show that the prediction using the mixture model of all neural network models in the model class is bayes optimal. However, this mixture model is difficult to calculate strictly for neural network models, since the complex integration on the parameter space is cannot be calculated for the general priors. We, therefore, propose the new prediction method with asymptotic Bayes optimality, based on Laplacian method which calculates the asymptotic posterior predictive distribution and then removes the integration, apply this method to the multilayer neural network models and verify the efficiency of the proposal through the simulation experiments.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Neural Network / Bayes Decision Theory / Mixture Model / Generalization
Paper # NC-95-121
Date of Issue

Conference Information
Committee NC
Conference Date 1996/3/18(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) On Bayes Optimal Prediction based on Mixture Model of Multilayer Neural Networks
Sub Title (in English)
Keyword(1) Neural Network
Keyword(2) Bayes Decision Theory
Keyword(3) Mixture Model
Keyword(4) Generalization
1st Author's Name Hiroki Hashikawa
1st Author's Affiliation Musashi Institute of Technology()
2nd Author's Name Masayuki Gotoh
2nd Author's Affiliation Waseda University
3rd Author's Name Nobuhiko Tawara
3rd Author's Affiliation Musashi Institute of Technology
Date 1996/3/18
Paper # NC-95-121
Volume (vol) vol.95
Number (no) 598
Page pp.pp.-
#Pages 8
Date of Issue