Back to the program.
Speeding Up Evolution
Jens Ziegler
University of Dortmund
Dept. of Computer Science
Systems Analysis


The high number of fitness evaluations in evolutionary algorithms is often expensive, time-consuming or otherwise problematic in many real-world applications. Reducing the number and duration of evaluations is thus a desirable goal in order to reduce the overall runtime of the evolutionary process. It can be shown that the result of a classifier, trained to discriminate between good and worse individuals, can be used in combination with tournament selection to both yield better results with less evaluations.


Evolutionary Algorithms, Classification, Meta-Evolution