Improvements to the design of interactive Evolutionary Multiobjective Algorithms (iEMOAs) are unlikely without quantitative assessment of their behaviour in realistic settings. Experiments with human decision-makers (DMs) are of limited scope due to the difficulty of isolating individual biases and replicating the experiment with enough subjects, and enough times, to obtain confidence in the results. Simulation studies may help to overcome these issues, but they require the use of realistic simulations of decision-makers. Machine decision-makers (MDMs) provide a way to carry out such simulation studies, however, studies so far have relied on simple utility functions. In this paper, we analyse and compare two state-of-the-art iEMOAs by means of a MDM that uses a sigmoid-shaped utility function. This sigmoid utility function is based on psychologically realistic models from behavioural economics, and replicates several realistic human behaviours. Our findings are that, on a variety of well-known benchmarks with two and three objectives, the two iEMOAs do not consistently recover the most-preferred points. We hope that these findings provide an impetus for more directed design and analysis of future iEMOAs.
|Title of host publication||GECCO '21|
|Subtitle of host publication||Proceedings of the Genetic and Evolutionary Computation Conference|
|Publisher||Association for Computing Machinery|
|Number of pages||9|
|Publication status||Published - 26 Jun 2021|
|Event||2021 Genetic and Evolutionary Computation Conference, GECCO 2021 - Virtual, Online, France|
Duration: 10 Jul 2021 → 14 Jul 2021
|Name||Genetic and Evolutionary Computation Conference (GECCO)|
|Conference||2021 Genetic and Evolutionary Computation Conference, GECCO 2021|
|Period||10/07/21 → 14/07/21|
Bibliographical noteFunding Information:
M. López-Ibáñez is a “Beatriz Galindo” Senior Distinguished Researcher (BEAGAL 18/00053) funded by the Ministry of Science and Innovation of the Spanish Government.
© 2021 ACM.
- Design of experiments
- Interactive evolutionary multi-objective optimization
- Machine decision maker
ASJC Scopus subject areas
- Computational Mathematics