 |
Evolutionary
algorithms are a family of algorithms in artificial intelligence
whose
strategy is to mimic evolution in biological systems to arrive at
solutions. As happens in nature, new solutions are arrived at by
various methods of combining or mutating existing solutions.
These new solutions are then used, in turn, to generate future
generations. As is in many cases of evolution, the process can be set
to allow more of the better, fitter solutions to propagate into future
populations. If a solution space can be imagined as a 3-D surface with
one solution peak, then this method may find that peak with some
efficiency. But, if there are multiple peaks in the solution space,
then just using the best solutions of a single generation may head
towards a false solution peak. To avoid this, non-optimal or even
random individuals may be left to propagate into future populations.
The randomness introduced by allowing a variety of individuals in a
generation to "reproduce" gives these types of algorithms the power to
find unexpected solutions. As an example, imagine a space rover. There
are many situations that the designer of that rover can predict and
design the rover to react in a proper way. But, if the rover is to go
into space, it may encounter problems which the designers had no
possible way to predict. In these cases, using an evolutionary
algorithm would allow the rover to come up with a new way to climb over
an object or collect a sample that was not hard-coded
into
its "thinking" hardware.
What this project brings to the researcher and developer of
evolutionary algorithms, is a quick, close to optimal way to put their
algorithms into hardware. When one thinks of algorithms, one thinks of
mathematicians and simulations in software. We give the scientists with
strengths in fields other than hardware a simple way to put their
algorithms into hardware and test them in the real world.
We hope that our testbed will accelerate research using evolutionary
algorithms and allow us to attempt solutions of problems that
would take years to examine and solve otherwise. |
 |
Preliminary results were
announced at ESDIS, a section of a computational intellegence
conference held March 30-April 2 2009 in Nashville. We will be pursuing
a beta plus version to be done by the end of summer 2009.
Our paper's name is: "An Evolutionary Algorithm Testbed for Quick Implementation of Algorithms in Hardware" and the authors are
Tina Smilkstein, Kiran Kumar Tati, Parashar Barve, M. Lutful Hai, Kittisak Sajjapongse and Durgesh K. Sharma. |
Flexible
Hardware for Quick Implementation of
Evolutionary Algorithms
Mizzou.edu
