We've gotten good at using computers
to supplement what human brains do poorly (e.g. numerical
integration), but brains and computers work very differently. In the
old science fiction movies, white-coated actors used to call the huge
boxes with whirling tape drives and myriad flashing lights an
'electronic brain,' but even then scientists knew that this was not
at all true. Since brains can perform some functions easily that
computers can do only with great difficulty - if at all - this begs
the question of how we an build computers which are more like brains.
Neural networks are a technology which has just
taken off in the last 15 years or so, and is starting to go
mainstream. In simple terms, neural nets attempt to crudely imitate
the biological processes for recognizing, classifying and responding
appropriately to external stimuli. This is sometimes called 'parallel
distributed processing.' The most sophisticated artificial neural
nets are a long way from even the smallest vertebrate brains, but
still show great promise for solving problems which are very
difficult to solve robustly with conventional solutions.
I'm interested in exchanging Matlab scripts or related
information pertaining to applications of neural nets. My own NN
projects so far have been primitive. Among other things, I am
interested in applications of NNs to smart spacecraft.
Links:
Articles
Novel
Neural Net Recognizes Spoken Words Better Than Human
Listeners.
The Learning Processor
Neural Network FAQ.
Neural Network Applications
A thesis on recurrent neural nets.
An article on NNs for the General Reader.
An
article on why brains and computers are different.
NN Sites
Teuvo
Kohonen's page.
About the work of this important Finnish
researcher. Most of the content is in English.
NeuroQuest.
Developers of a Linux-based NN toolbox.
NEuroNet.
At King's College in London.
Demo GNG.
A very cool site which uses Java to demonstrate the
'Neural Gas' concept.
The
Backpropagator's Online Reading List.
Evolutionary
Development of Neural Architectures
At Iowa State University. Combining Neural nets and
genetic algorithms sounds like a great idea to me. There is also a
companion
paper.
Neural
Net Free Software FAQ.
Very useful.
Neural
Net Software
A list of available software for neural nets. This
is a European site, so some of the software listed may be hard to
come by stateside. An interesting site nevertheless.
Center for Theoretical
Neuroscience
At Baylor College of Medicine. These folks are
working on some neat stuff. If so inclined, you can go get your
MD/PhD at Baylor, and work on neuroscience.
NN Applications Sites
Terra
Research.
A NN software company.
University
of Glasgow.
Describes research in Neural adaptive control.
Computational
Sciences Division
at NASA Ames has some cool Neural Net projects
summarized here.
Parallel Distributed Processing in General
Big Science.
A very cool idea. Check them out.
I think
genetic algorithms - a subset of 'population based methods' are even
more fascinating than neural nets, and not only because they hold
vast promise for solving all kinds of problems.
Again, we're trying to imitate one of nature's
great successes to achieve globally optimized solutions to problems
which don't depend on the imagination or cleverness of humans. After
all, humans can really only be really creative about items in the
immediate consciousness - usually about 7 things for most people. The
really cool thing is that great imagination can be built into an
algorithm using some simple principles with familiar biological
analogies, and this is not limited to what human consciousness can
handle. A genetic algorithm can find
solutions which require finely balancing a large number of
independent variables - something nature does all the time.
General Evolutionary Computation Sites
A very cool Java applet illustrating genetic
algorithms. Interesting thingds happen when you try to fool it.
A brief description and bibliography.
This is a for-profit company which
applies natural computation to real problems.
At the Sante Fe
Institute
This is probably the only site you
need. It has a wide selection of links and lots of other useful
information.
What more can you ask for? Humor
and EC in one action packed site.
Bills itself as the "Hitchhiker's
Guide to Evolutionary Computation."
offered by Johns Hopkins University.
An artistic application of genetic
algorithms. A visual treat.
Papers and Articles
July 1999.
April 1999.
A 1999 article in the Santa Fe
Institute Bulletin about a researcher who uses Genetic Algorithms to
critique economic models.
Most genetic algorithms are
haploid, whereas most sexually reproducing creatures are diploid or
even mutiploid. My very dry review of some of the literature, and a
few comments of my own.
Tutorials and other Educational Resources
A Book on DNA Computing.
L-Systems Tutorial
This is a brief, easy to understand
explanation of Lindenmayer systems, which use simple rules to - among
other things - describe the growth of plants. Jari Vaario has used L
systems in his research.
Software and Code
Simon Frasier's
MacTierra page.