Evolution of Artificial Neural Networks |
In recent years, computer scientists have been trying
to develop artificial neural networks for use in directing robots.
While these artificial networks are obviously much simpler than the human
brain, they are based on the same ideas. Traditional networks model
the use nodes which are connected by wires and imitate the firing of a
synapse in the brain. One of the areas within this field is the development
of evolved networks. A number of possible networks are generated,
tested for fitness based on predetermined criteria, and then either mutated
or allowed to "reproduce" based on their fitness. With the discovery
of NO as a neurotransmitter, a group at the University of Sussex is expanding
these ideas to include the possibility of the concentration of a gas affecting
the output of a node. The inclusion of a virtual gas adds a flexibility
to the system which significantly lowers the number of generations required
to reach a functional network.
Okay, so there isn't much chemistry here, but it's
fun stuff anyway. Read the "Gas on the Brain" paper that we got at
the beginning of the semester (if you've still got it) for the general
idea. If you're desperate to see my presentation right now, it's
on my homepage.
The most relevant paper I have is in the References section. It's
called "Better Living Through Cemistry: Evolving GasNets for Robot Control,"
and you can find it here.
See ya Wednesday. Kelda