Paper Abstract and Keywords |
Presentation |
2010-03-11 17:15
Analysis of Autocorrelation type Associative Memory Model with Hierarchical Patterns by Using PCA Teijiro Shiotsuka (Waseda Univ.), Kenji Nagata, Koji Hukushima (Tokyo Univ.), Masato Okada (Tokyo Univ./RIKEN), Masato Inoue (Waseda Univ.) NC2009-158 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
The statistical mechanical approach is useful and has been applied to various problems in the field of information processing, but its application is mainly limited to the class of the mean-field models. This report proposes a more general analysis method and validates its effectiveness. This method combines exchange Monte Carlo method and principal component analysis (PCA), and visualizes obtained high-dimensional empirical distribution. Through this method, we try to estimate the macro state of a given system. We also estimate an effective number of principal components by introducing probabilistic principal component analysis (PPCA). We have applied this approach to associative memory model, and obtained consistent results with those by the mean-field analysis. Those results suggest the possibility that we can analyze not-analytically-solved systems by using this approach. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
statistical mechanics / exchange Monte Carlo method / principal component analysis / probabilistic principal component analysis / autocorrelation type associative memory model / hierarchical patterns / / |
Reference Info. |
IEICE Tech. Rep., vol. 109, no. 461, NC2009-158, pp. 413-418, March 2010. |
Paper # |
NC2009-158 |
Date of Issue |
2010-03-02 (NC) |
ISSN |
Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
NC2009-158 |
|