Towards Automatic Testing of Reference Point Based Interactive Methods

Abstract

In order to understand strengths and weaknesses of optimization algorithms, it is important to have access to different types of test problems, well defined performance indicators and analysis tools. Such tools are widely available for testing evolutionary multiobjective optimization algorithms. To our knowledge, there do not exist tools for analyzing the performance of interactive multiobjective optimization methods based on the reference point approach to communicating preference information. The main barrier to such tools is the involvement of human decision makers into interactive solution processes, which makes the performance of interactive methods dependent on the performance of humans using them. In this research, we aim towards a testing framework where the human decision maker is replaced with an artificial one and which allows to repetitively test interactive methods in a controlled environment.