Initial Population Diversity Does Not Influence Performance

- It is widely believed that greater initial population diversity leads to improved performance in genetic algorithms. However, this assumption has not been rigorously tested previously. We put this assumption to the test on two benchmark problems and found that greater diversity did not lead to improved performance. This result will require a serious rethinking on the part of the evolutionary computation community as to why genetic algorithms sometimes perform very differently on successive applications to the same problems.

• Type: Conference
• Year: 2008
• Where: GEM
• Authors: Pedro A. Diaz-Gomez, Dean F. Hougen
• Keywords: Artificial Intelligence, GEM 2008, Genetic Algorithms, Greater Initial Population, Improved Performance