Probabilistic Decision Making for Interactive Evolution with Sensitivity Analysis
Jonathan Eisenmann, Matthew Lewis, Rick Parent: Probabilistic Decision Making for Interactive Evolution with Sensitivity Analysis. In: EvoMUSART 2014, S. 1-12.
Recent research in the area of evolutionary algorithms and interactive design tools for ideation has investigated how sensitivity analysis can be used to enable region-of-interest selection on design candidates. Even though it provides more precise control over the evolutionary search to the designer, the existing methodology for this enhancement to evolutionary algorithms does not make full use of the information provided by sensitivity analysis and may lead to premature convergence. In this paper, we describe the shortcomings of previous research on this topic and introduce an approach that mitigates the problem of early convergence. A discussion of the trade-offs of different approaches to sensitivity analysis is provided as well as a demonstration of this new technique on a parametric model built for character design ideation.