Single-criterial global optimization via Evolutionary Algorithms:
EAs are generally not pure locally acting optimization algorithms. To a
certain (controllable!) extend, they are also well suited for multimodal
problems. Up to now, systematic investigation of this property has
been largely neglected. I am interested in evaluating the possibilities
and limits for EAs on these problem types. This includes approaches
like Niching and Multistarts.
Experimental analysis of Evolutionary Algorithms:
Recent methods as the Sequential Parameter Optimization (SPO)
by Thomas Bartz-Beielstein permit to locate good algorithm
parameter sets for treating a concrete problem. Additionally,
experimental data may be employed to obtain further insight:
How do the parameters interact, and how difficult is it to
adapt an optimization algorithm to a certain problem?
Computational Intelligence in Games:
Here, I am mostly interested in the design of
new user-centric extensions of realtime strategy
games and the construction/evolution of believable
non-player characters (NPCs) to enhance player satisfaction. Moreover,
the setting is a very difficult test environment for
optimization/learning algorithms: High system dynamics,
mutiple criteria, hard time constraints....
Multicriterial Evolutionary Algorithms:
Here I am most interested in the interrelation of Pareto set
and Pareto front. Motivated by the multimodal viewpoint
onto singleobjective problems we find interesting similarities:
There is no reason to believe that the single target functions
of multicriterial problems are always unimodal. However, this
is an implicit assumption of most current multicriterial
Evolutionary Algorithms.
Real-World applications of Evolutionary Algorithms:
Preferably, I am dealing with applications of EA on problems
from the following domains: scheduling, classification, map labeling,
thermodynamical process engineering, and
elevator control.