LATERALLY INTERCONNECTED SELF-ORGANIZING FEATURE MAP IN
HANDWRITTEN DIGIT RECOGNITION
Yoonsuck Choe and Risto Miikkulainen
Under Construction
Abstract
An application of biologically motivated laterally interconnected
synergetically self-organizing maps (LISSOM) to off-line recognition of
handwritten digit is presented. The lateral connections of the LISSOM map
learns the correlation between units through Hebbian learning. As a result,
the excitatory connections focus the activity in local patches and lateral
connections decorrelate redundant activity on the map. This process forms
internal representations for the input that are easier to recognize than the
input bitmaps themselves or the activation patterns on a standard
Self-Organizing Map (SOM). The recognition rate on a publically available
subset of NIST special database 3 with LISSOM is 4.0% higher than that
based on SOM, and 15.8% higher than that based on raw input bitmaps.
These results form a promising starting point for building pattern recognition
systems with a LISSOM map as a front end.
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
| (a)Input Samples |
(b)Initial Response |
(c)Settling |
(d)Final Response |
(e)Scale |
Figure 1: The settling of map activity through repeated, recursive lateral
interaction with digits 5 (upper row) and 8 (lower row) are shown.
(a) The two input samples, 5 and 8. (b)
The initial response of the map shows wide spread and unfocused activity.
(c) The animation shows the settling of the map activity through
recursive lateral interaction. (d)The final response of the map
after 15 settling iterations shows focused and sharp activity. Also, it is
notable that the initial response of the map to the two different digits
5 and 8 are similar, the settled activity becomes distinct from each
other.(e) The color scale codes the activity from 0.0 (black)
through 1.0 (white).
Related Publications
- Yoonsuck Choe, Joseph Sirosh, and Risto Miikkulainen.
"Laterally Interconnected
Self-Organizing Maps in Hand-Written Digit Recognition" , To appear
in
D.S.Touretzky, M.C.Mozer, and M.E.Hasselmo. (Editors), Advances in
Neural Information Processing Systems 8, pages 736-742.
MIT Press, Cambridge, MA, 1996.
- Yoonsuck Choe. "Laterally
Interconnected Self-Organizing Feature Map in Handwritten Digit Recognition",
Techical Report AI95-236, Department of Computer Sciences,
University of Texas at Austin, August 1995 [Masters Thesis]
This page is maintained by Yoonsuck Choe (yschoe@cs.utexas.edu).
Last Updated :
Mon Sep 29 03:54:46 CDT 1997