Date: Thu, 07 Nov 1996 19:12:07 GMT
Server: NCSA/1.5
Content-type: text/html
Last-modified: Fri, 01 Nov 1996 04:24:49 GMT
Content-length: 6989
David J. Finton's Home Page
David J. Finton
finton@cs.wisc.edu
Computer Sciences Department
University of Wisconsin-Madison
1210 West Dayton Street
Madison, WI 53706
(608) 262-9275
Welcome to my page! I'm a grad student / research
nerd
in artificial
intelligence
here at the
University of Wisconsin-Madison.
I grew up in
Grand Rapids, Michigan,
(which is the
Late Show's
ex-Home Office),
earned a degree in
math at Michigan State,
and a master's in
computer science
here at the UW.
I'm now a dissertator at this
institution,
after taking a little over a year to develop traffic measurements software
for
AT&T
after my first thesis advisor left Wisconsin.
When I'm not at my trusty
NeXTstation
or the library, I enjoy
playing trumpet and piano, listening to
"longhair music",
playing volleyball with the
InterVarsity
folks, and contributing to the SuperSoaker arms race.
If you have any comments about my pages, feel free to use my
comment form,
or just send me e-mail.
Or finger my account
to see my current plan and whether I'm on the system.
Gainful employment:
I am a TA for
CS 540,
Introduction to Artificial Intelligence.
Current Project:
If computers are so smart, why do we have to understand them?
Making machines more
intelligent
is the goal of Artificial Intelligence. To me, the essence of intelligence
is the ability to learn and adapt, to learn to act
appropriately in order to reach our goals.
Reinforcement learning treats this problem in the general case where
the system has outputs to control actions that can change its environment,
and it has inputs through which it senses its environment. It also has an input
for reinforcement, which is a weak kind of feedback which can be expressed as
a positive or negative number. So, instead of having a teacher to present the
system with input/output pairs, the system instead receives "thumbs up" or
"thumbs down" at irregular intervals.
My work has focussed on how the need to
distinguish good actions from bad ones can direct the process of building a
good representation of the environment in terms of relevant, or
important features. (See my note on
importance-based
feature extraction). Currently I am applying this notion of
importance to the problem of learning to balance the need to
explore the world with the need to perform optimally (exploration vs. exploitation).
I am also investigating ways of using importance to make the learning
process more efficient by allowing the system to specify the starting points for
its learning experiments
(active learning). My goal is to develop a better understanding of intelligent
adaptation. I hope that this will provide a basis for intelligent action which
will also benefit from knowledge-based and task-based work. See my (really
out-of-date, sorry!)
reinforcement learning page
for more information.
My Hotlist
This is
my browser-independent hotlist. I keep a copy here so I can
access it from any of the browser/platform combinations I use.
It's actually my Bookmarks file from
OmniWeb,
which is a
more elegant and more functional browser than Netscape, in my opinion. OmniWeb
is currently only available for NEXTSTEP, but will be available for
all the
OpenStep
variants when OpenStep is released.
My Editorial Pages:
My
response
to the Jehovah's Witnesses on the deity of Christ
Wisconsin Sites:
Some of My Favorite Places to Visit:
Last modified: October 31, 1996
finton@cs.wisc.edu