For best experience please turn on javascript and use a modern browser!

Marleen de Jonge has published a part of her BSc-thesis, even before finishing the project, as a Late Breaking Abstract on the top-knotch conference EvoStar2020.

Evolutionary algorithm

"We have tested the impact of different parameter settings on the performance of the plant propagation algorithm. It's an evolutionary algorithm that performs well on a lot of complex problems, such as the Traveling Salesman Problem and University Timetabling, but we don't know how its parameters should be tuned for optimal results. It's a problem typical for algorithms in the field of evolutionary computation; new algorithms are proposed with great pace, but the impact of parameter settings is hardly tested.

Mutually invariant

The strength of our current findings is not only that we identified a window of optimal values within 400 different parameter settings, but also that they are largely mutually invariant. In other words: just being inside the window is enough, practically erasing two parameters from the algorithm", says Marleen.

"It's a good find, especially because of its consistency, but we really have to see how well these results hold up when testing against a broader range of problems and instances. Having said that, the Plant Propagation Algorithm has shown a number of advantages recently, so our confidence steadily grows." says Daan van den Berg, Marleen's supervisor.

The conference, scheduled for Sevilla, was completely held online.