Evolutionary Optimization in Biology and Numerics by Abstraction and Synthesis of Natural Organisms" Behavior
EvolutionFrom 11/1998 to 10/2001
Principal Investigator: Alois Kastner-Maresch
Staff: Clemens Frey
The project touches a variety of neighboring disciplines.
Especially numerical mathematics and a kind of abstracted evolutionary biology
have effects on the project. Evolutionary processes are synthesized on
simple organism which have abstract abilities
(like eating, growing, producing offspring ...). These abilities will be
governed by simple rule sets (like if-clauses or Lindenmayer-systems) and by
the (phenotypical) struggle for food.
Food (or energy) is the preliminary for the individuals taking different actions.
During evolution on these rule-sets different species will show up. They will
compete for food on artificial landscapes, and become species in different
niches of the landscapes. Because they can grow, they search food in
different regions, implying that by knowing a fast way to globally good regions
of the landscape makes an individual perform well in the competition against
the other members of a population.
Going on in this metaphor, individuals should emerge performing well as
global optimizers of some target functions. And this is the point where to return to numerical mathematics.
This kind of Evolutionary Algorithm (Genetic Programming on Syntactical Structures)
works on a meta level. It will not optimize on its own, it evolves optimizing
individuals. This process as well as the optimizing performance of the resulting
individuals have to be examined.
Remarkable about the Evolutionary Process is the fact that we will try to
exclude any central "god" who has an exogeneous target function in his hands
measuring the optimizing individuals. Local competition will be the key for that.
|