Thesis available: Transfer between neural networks

[Lorien Y. Pratt]

FTP-host: archive.cis.ohio-state.edu
FTP-file: pub/neuroprose/pratt.thesis.ps.Z

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

Transferring Previously Learned Back Propagation Networks to New Learning Tasks
144 pages
Lorien Y. Pratt
Colorado School of Mines (Dissertation written at Rutgers University)

ABSTRACT: When people learn a new task, they often build on their
ability to solve related problems.  For example, a doctor moving
to a new country can use prior experience to aid in diagnosing
patients.  A chess player can use experience with one set of
end-games to aid in solving a different, but related, set.  However,
although people are able to perform this sort of skill transfer
between tasks, most neural network training methods in use today
are unable to build on their prior experience.  Instead, every new
task is learned from scratch.

This dissertation explores how a back-propagation neural network
learner can build on its previous experience.  We present an
algorithm, called Discriminability-Based Transfer (DBT), that
facilitates the transfer of information from the learned weights
of one network to the initial weights of another.  Through evaluation
of DBT on several benchmark tasks we demonstrate that it can speed
up learning on a new task.  We also show that DBT is more robust
than simpler methods for transfer.

--------
Sample session for obtaining this file via anonymous ftp:
> ftp archive.cis.ohio-state.edu
Name: anonymous
Password: (type your email address her)
ftp> cd pub/neuroprose
ftp> binary
ftp> get pratt.thesis.ps.Z
ftp> quit
> uncompress pratt.thesis.ps.Z
> lpr pratt.thesis.ps

---------
For this with email but no ftp access, you can use the mail server at Rutgers,
where this document is ML-TR-37.

Send email to mth at cs.rutgers.edu to request the document, or send the word
`help' in a message to:
		   ftpmail at decwrl.dec.com

---------
Those without any network access at all can receive copies of this document by
requesting ML-TR-37 from the following address:
	  Technical Reports Librarian
	  Computer Science Department
	  Rutgers University
	  New Brunswick, NJ 08903
	  USA
	  [Currently there is no charge.]


Dr. L. Y. Pratt                       Dept. of Math and Computer Science
lpratt at mines.colorado.edu             Colorado School of Mines    
(303) 273-3878 (work)                 402 Stratton                
(303) 278-4552 (home)                 Golden, CO 80401, USA      

Note: I'll be travelling out of the country for the next month.  If you have
trouble printing this paper, I'll be available to help after I return on July
16.  Send me email -Lori