Abstract
The main target of multi-objective evolutionary approaches is to find a set of well-distributed compromising solutions that precisely approximate the entire Pareto front. Although there are various evolutionary strategies for solving multi-objective optimization problems, however, they can be computationally expensive for complex practical problems. Moreover, a reasonable number of solutions should be provided for the decision maker so that he/she could make an adequate decision avoiding complex analysis of large amount of information. Therefore, optimization methods based on the decision maker’s preferences are useful in which the region of interest of the Pareto front is approximated. We have developed a new synchronous R-NSGA-II algorithm, where concepts of the interactive synchronous NIMBUS method are combined with the preference-based evolutionary R-NSGA-II algorithm. In the proposed algorithm several scalarizing functions are used simultaneously, therefore the several sets of grouped solutions are obtained from the same preference information.
When we deal with complex problems, interactive multi-objective optimization methods should be used. Naturally, interactive methods incorporate other types of multi-objective approaches as well as evolutionary algorithms. In the interactive methods DM provides preference information progressively during the solution process and obtains solutions derived based on this information as feedback. In another investigation, we have propose the new visualization technique based on dimensionality reduction for enhancing interactive methods. It uses the multi-dimensional scaling method, that is grounded on the preservation of the similarity between Pareto front points when their dimensionality is reduced. We have realized the technique as an independent tool, and integrated it into IND-NIMBUS decision support system.
Biography
Ernestas Filatovas received his PhD at the Vilnius University, Lithuania in 2012.
He is currently a post-doctoral researcher at the Vilnius University Institute of Mathematics and Informatics. His main research interests include multi-objective optimization, multi-objective evolutionary algorithms, multiple criteria decision making, parallel computing, data visualization, and image processing.