Solving high objective problems in fixed interactions with the decision maker
Abstract
In recent advancements towards handling high objective optimization problems, it is proposed to progressively integrate the decision maker with the execution of an evolutionary multi-objective optimization algorithm. Preferences from the decision maker are accepted at the intermediate steps of the algorithm and a progress towards the most preferred point is made. In this paper, we extend the work on `progressively interactive evolutionary multi-objective optimization using value function' (PI-EMO-VF) by allowing the optimization to be performed in a fixed number of interactions with the decision maker. In the PI-EMO-VF procedure, information is accepted from the decision maker, which is utilized by the evolutionary algorithm to perform a focused search in the region of interest. However, it is not possible to restrict the number of interactions required to handle an optimization problem. This paper contributes towards, solving the optimization problem in a pre-decided number of decision maker calls. Once the available budget of decision maker calls are known, it is optimally utilized to get close to the most preferred point on the Pareto-frontier. The paper evaluates the performance of the modified PI-EMOVF algorithm on two, three and five objective test problems. A comparative study is performed against the previous proposal for the PI-EMO-VF procedure.
Authors: Ankur Sinha, Anmol Pandey, Kalyanmoy Deb
Published in: IEEE Congress on Evolutionary Computation (CEC) (2012)