No subject

[Tom Fuller]

The file fuller.thesis.ps.Z is now available for copying from the 
Neuroprose repository:

Supervised Competitive Learning: a technology for pen-based adaptation in real time

Thomas H. Fuller, Jr.
Computer Science
Principia College
Elsah, IL 62028
tom at csc1.prin.edu

Abstract:

The advent of affordable, pen-based computers promises wide application in 
educational and home settings. In such settings, systems will be regularly 
employed by a few users (children or students), and occasionally by other 
users (teachers or parents). The systems must adapt to the writing and 
gestures of regular users but not lose prior recognition ability. 
Furthermore, this adaptation must occur in real time not to frustrate or 
confuse the user, and not to interfere with the task at hand. It must also 
provide a reliable measure of the likelihood of correct recognition.

Supervised Competitive Learning is our technology for the recognition of 
handwritten symbols.  It uses a shifting collection of neural network-based
similarity detectors to adapt to the user. We demonstrate that it satisfies 
the following requirements:

1.  Pen-based technology: digitizing display tablet with pen.
2.  Low cost: PC-level processor with about 50 MIPS.
3.  Wide range of subjects: varying by age, nationality, writing style.
4.  Wide range of symbol sets: numerals, alphabetic characters, gestures.
5.  Usage: adaptation to regular users; persistent response to occasional users.
6.  On-line recognition:  both response and adaptation in real time.
7.  Self-criticism: reliable measure of likelihood of correct response.
8.  Context-free classification: symbol by symbol recognition.

SCL successfully recognizes handwritten characters from writers on which 
it has trained (digits, lowercase, uppercase, and others) at least as well 
as known current systems (96.5% - 99.2%, depending on character sets).  
It adapts to its user in real time with a 50 MIPS processor without loss 
of response to occasional users.  Finally, its estimates of its correctness 
are strongly correlated with actual likelihood of correctness.

This is a doctoral dissertation at Washington University in St. Louis. Hardcopies 
are only available from University Microfilms, Inc. This work was supported by the 
Kumon Machine Project.

ADVISOR: Professor Takayuki Dan Kimura
completed December, 1994

Department of Computer Science
Washington University
Campus Box 1045
One Brookings Drive
St. Louis, MO  63130-4899

queries about the work should go to;

Thomas H. Fuller, Jr.
Computer Science
Principia College
Elsah, IL 62028
tom at csc1.prin.edu


Here's a sample retrieval session:

unix> ftp archive.cis.ohio-state.edu
Connected to archive.cis.ohio-state.edu.
220 archive FTP server (Version wu-2.4(2) Mon Apr 18 14:41:30 EDT 1994) ready.
Name (archive.cis.ohio-state.edu:me): anonymous
331 Guest login ok, send your complete e-mail address as password.
Password: me at here.edu
230 Guest login ok, access restrictions apply.
Remote system type is UNIX.
Using binary mode to transfer files.
ftp> cd pub/neuroprose/Thesis
250 CWD command successful.
ftp> get fuller.thesis.ps.Z
200 PORT command successful.
150 Opening BINARY mode data connection for fuller.thesis.ps.Z (798510 bytes).
226 Transfer complete.
798510 bytes received in 180 seconds (5 Kbytes/s)
ftp> bye
221 Goodbye.
unix> uncompress fuller.thesis.ps.Z
unix> <send fuller.thesis.ps to favorite viewer or printer>