Illinois Genetic Algorithms Laboratory

A while ago, Christoph, a visiting student at the product development research lab was talking to me about how he was having some problems with solving certain instances of his seach problem. He told me that the problematic GA runs would converge to a mediocre solutions, but different parts within those solution were of good quality. I remembered Clarissa’s work—that she did during her visit to IlliGAL—about how strong symmteries in problems such as Ising and other spin-glass models can cause synchorinization problems and how niching methods can be used to avoid them. So, I suggested him that there might be synchronization issues with the problematic instances and using a niching method might be the solution. Lo and behold, he came back yesterday and said using niching significantly improved the GA performance!

Clarissa Van Hoyweghen’s Ph.D. Thesis Symmetry in the Representation of an Optimization Problem and related papers, available as IlliGAL technical reports 2001020 and 2001030 are a good starting point for anyone interested in knowing more about symmetry and synchronization problems in optimization in general, and GAs in particular.