Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!bloom-beacon!husc6!bunny!rich
From: rich@bunny.UUCP (Rich Sutton)
Newsgroups: comp.ai.neural-nets
Subject: Re: Training
Message-ID: <6720@bunny.UUCP>
Date: 23 Mar 89 16:17:59 GMT
References: <2698@sun.soe.clarkson.edu>
Reply-To: rich%gte.com@relay.cs.net (Rich Sutton)
Organization: GTE Laboratories, Waltham, MA
Lines: 35

In article <2698@sun.soe.clarkson.edu> spam@clutx.clarkson.edu writes:

> Has anyone come up with a good way to train a net
> without knowing a "target" in advance?
...
> All the learning methods I've studied so far
> are inadequate for this.

The learning methods that you are looking for are called
"reinforcement learning" methods.  They are less well known than
supervised learning methods, but hardly obscure.  And of course they
are more powerful in the sense you refer to -- they can learn what to
do without a teacher that already knows the correct answers.  
I recommend the following papers.
					-Rich Sutton


A. Barto, R. Sutton, & P. Brouwer, ``Associative search network:
A reinforcement learning associative memory," Biological
Cybernetics, 40, 1981, pp. 201--211.

A. Barto & R. Sutton, ``Landmark learning: An illustration of
associative search," Biological Cybernetics, 42, 1981, 
pp. 1--8.

Barto, Sutton, and Anderson, ``Neuronlike elements that can solve
difficult learning control problems,'' IEEE Trans. on Systems, Man, and
Cybernetics, 13, 1983, pp. 835--846.

Williams, R. J., ``Toward a theory of reinforcement-learning
connectionist systems,'' Technical Report NU-CCS-88-3, College of
Computer Science, Northeastern University, Boston, Massachusetts, 1988.

My dissertation is also directly relevant and I invite you to write me
for a copy at GTE Labs, Waltham MA 02254.