Yafir Snir- Knowledge Transfer in Evolutionary Search of Conceptual Path Plans

Electrical Engineering Systems Seminar

Speaker: Yafir Snir

M.Sc. student under the supervision of Dr. Amiram Moshaiov

Wednesday, 31^st January 2024, at 15:00

Room 011, Kitot Building, Faculty of Engineering

Knowledge Transfer in Evolutionary Search of Conceptual Path Plans

Abstract

This study is motivated by the need to develop efficient Evolutionary Multi-Concept Optimization (E-MCO) tools. Such computational search tools aim to support the selection of a conceptual solution (concept) by decision-makers under conflicting objectives.  In the considered search techniques, a concept is viewed as a meaningful set of solutions which have some common features. We address the challenge of improving E-MCO tools by exploring the utilization of knowledge transfer approaches. Past studies on E-MCO evolved the considered concepts with no knowledge transfer between the evolved concepts, and in particular no crossover was used between solutions of different concepts. In contrast, this research investigates the potential benefits of eliminating the crossover restriction among solutions from different concepts, aiming to facilitate fast convergence to the true Pareto front. Various knowledge transfer approaches are suggested based on a probabilistic model of the solution space and an online-learned random mating probability matrix. First, we develop a benchmark suite of conceptual path planning problems. Next, we evaluate the proposed transfer techniques on this benchmark test suite. The obtained results indicate that knowledge transfer enhances performance for most cases, revealing novel solutions. However, challenges such as asymmetric effects highlight the need for further research to optimize knowledge transfer mechanisms per concept.

השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בדף הנוכחות שיועבר באולם במהלך הסמינר