Approaching epistemic and aleatoric uncertainty with evolutionary optimization: examples and challenges

Josu Ceberio, Juan Carlos Cortés, Francisco Fernández de Vega, Oscar Garnica, José Ignacio Hidalgo, J. Manuel Velasco, Rafael-Jacinto Villanueva

Abstract

Uncertainty quantification is an emerging area in theory and applications. There are various approaches for modeling and dealing with uncertainty. However, when modeling real-world problems, the calibration considering uncertainty is a relevant issue on which little has been studied. We think that the applicability of evolutionary computation can be much more significant when dealing with uncertainty due to the capabilities of these kinds of algorithms to find suitable solutions in reasonable time budgets. In this context, evolutionary computation approaches have not been investigated extensively, except for a few papers that use metaheuristics and evolutionary algorithms to calibrate models with uncertainty successfully. This paper aims to motivate researchers to study and propose evolutionary algorithms to calibrate models with uncertainty in real-world problems and investigate new proposals. Uncertainty is present in data, measuring processes, evaluation, and models; hence it offers real challenges for the Evolutionary Computation community to propose more sophisticated algorithms and methods. In this paper, we make a general review of how uncertainty quantification has been treated in mathematical modeling, provide ideas on how to interpret uncertainty in specific problems, highlight its drawbacks and show two case studies in medicine.
BibTeX
@inproceedings{ceberio2022approaching,
  author = {Josu Ceberio and Juan Carlos Cortés and Francisco Fernández de Vega and Oscar Garnica and José Ignacio Hidalgo and J. Manuel Velasco and Rafael-Jacinto Villanueva},
  title = {{Approaching epistemic and aleatoric uncertainty with evolutionary optimization: examples and challenges}},
  booktitle = {GECCO Companion},
  year = {2022},
  doi = {10.1145/3520304.3533978},
}

Loading related content...

Figure