Constructing Hierarchical Neural Nets Using Sparse Distributed Memory

Type

conference paper

Language

english

Authors

Grebeníček František, Ing. (FEECS BUT)

URL

http://www.fit.vutbr.cz/~grebenic/Publikace/asis2000.ps.zip ZIP

Keywords

Neural nets, associative memory, Sparse Distributed Memory, pattern recognition

Abstract

This paper discusses a possibility of the hierarchical neural nets construction using Kanerva's Sparse Distributed Memory (SDM). SDM is an associative neural memory and can be used in visual pattern recognition. The paper introduces a hierarchical net for digit recognition. Results of xperiment show notable properties of the net: insensivity to digit position and warping. Finally, a possible modification of Fukushima's Neocognitron is discussed.

Annotation

This paper discusses a possibility of the hierarchical neural nets construction using Kanerva's Sparse Distributed Memory (SDM). SDM is an associative neural memory and can be used in visual pattern recognition. The paper introduces a hierarchical net for digit recognition. Results of xperiment show notable properties of the net: insensivity to digit position and warping. Finally, a possible modification of Fukushima's Neocognitron is discussed.

Published

2000

Pages

359-364

Proceedings

ASIS 2000 Proceedings of the Colloquium

Conference

22nd International Colloquium ASIS 2000 Advanced Simulation of Systems, Sv. Hostýn, Bystřice pod Hostýnem, CZ

ISBN

80-85988-51-8

Place

Sv. Hostýn, Bystřice pod Hostýnem, CZ

BibTeX

@INPROCEEDINGS{FITPUB5791,
   author = "Franti\v{s}ek Greben\'{i}\v{c}ek",
   title = "Constructing Hierarchical Neural Nets Using Sparse Distributed Memory",
   pages = "359--364",
   booktitle = "ASIS 2000 Proceedings of the Colloquium",
   year = 2000,
   location = "Sv. Host\'{y}n, Byst\v{r}ice pod Host\'{y}nem, CZ",
   ISBN = "80-85988-51-8",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/5791"
}