Patryk Filipiak, a junior lecturer at the Institute of Computer Science, University of Wroclaw (Poland), visited the CIplus research center in March. He is finalizing his PhD thesis concerning the behavior of “Proactive Evolutionary Algorithms in the Dynamic Constrained Optimization Problems” (DCOPs).
He proposed anticipation strategies that can speed up a convergence of candidate solutions in a class of DCOPs and applied these strategies to inverse kinematics problems. Patrick gave an interesting talk in SPOTSeven’s doctoral seminar. The beginning of his talk is shown in the following short video: http://www.gm.fh-koeln.de/~bartz/Videos.d/filipiak.m4v
The abstract of Patrick’s talk reads as follows:
Evolutionary Algorithms that aim at solving Dynamic Optimization Problems need to continuously explore the search space looking for the new optima and simultaneously they have to trace the ones that were found so far. It is typically achieved by applying a reactive mechanism that keeps re-evaluating certain specimens within the population in order to detect the possible landscape changes. However, the algorithms equipped with such mechanism are always one step back with the dynamic environment since they can only detect the changes that already happened. A proactive approach alleviates that issue by exploiting a dedicated forecasting model that anticipates the future landscape based on the historical data. As a result, the Evolutionary Algorithms can get ready for the changes to come (e.g. by directing some individuals into future promising regions). During my talk I would like to present the three prediction strategies namely – anticipation of evaluations, anticipation of optima locations and anticipation of landscape changes. Each of them guarantee a fast adaptability to certain types of dynamic environments by realizing the proactive approach.