khaosresearch / admpso

Artificial Decision Maker based on PSO

In the last decade, a series of interactive multi-objective optimizers based on reference point have been appearing to enable decision makers to dynamically include preference information in the learning procedures of such algorithms. However, when analyzing their performances, a critical issue is how to involve human decision makers into the interactive generation of reference points. This makes their performances to be highly dependent on the behavior of humans using them. A recent approach to this problem is to replace human decision maker with artificial one, to systematically evaluate and compare interactive multi-objective optimizers in controlled experimental procedures. In this study, a new artificial decision maker mechanism is proposed, which reuses the dynamics of particle swarm optimization for guiding the generation of consecutive reference points, hence simulating the decision maker in preference articulation procedures. The proposed approach is tested on the DTLZ benchmark with 3, 5, and 7 objectives, and using R-NSGA-II and WASF-GA as base reference point interactive optimizers. The experimental results show that the proposed approach is useful and efficient, in comparison with the previous one. It offers an intuitive and flexible mechanism to capture the current context in interactive decision making processes, when tacking multi-objective problems.

The code is available on https://github.com/jMetal/jMetal/tree/master/jmetal-core/src/main/java/org/uma/jmetal/util/artificialdecisionmaker