Probabilistic Decision Making for Interactive Evolution with
Sensitivity Analysis
J Eisenmann / Matt Lewis / Rick Parent
'Vast' character model
preview
preview
interactive evolutionary algorithms for the design of digital assets
typical interaction cycle
-
display
-
select
-
reproduce
-
repeat until satisfied
interactive design with
evolutionary algorithms
-
simple interface
-
creative support
-
subjective
-
human evaluation bottleneck
What if designers were given
the option to choose only
the parts of phenotypes
that they like?
inverse mapping with sensitivity analysis
sensitivity analysis with
the Morris method
sensitivity analysis with
the Morris method
Premature Convergence
sensitivity analysis with
the Sobol method
Morris heuristic
practical considerations
- Morris requires fewer model evaluations
- Sobol measures quantitative sensitivities more accurately
- Important variables
- model evaluation time
- number of region-of-interest selections
- number of model evaluations
- number of genes
crossover operator
crossover comparison
mutation operator
mutation comparison
design session comparison
design session comparison
design session comparison
design session comparison
design session comparison
design session comparison
design session comparison
design session comparison
design session comparison
Questions?
J Eisenmann
jeisenma@accad.osu.edu
accad.osu.edu/~jeisenma
Cara Malek / Matt Lewis
J Eisenmann / Lindsay Beach / Matt Lewis
Matt Lewis
J Eisenmann / Matt Lewis
J Eisenmann / Matt Lewis
Probabilistic Decision Making for Interactive Evolution with Sensitivity Analysis J Eisenmann / Matt L ewis / Rick Parent