A Practicebased Approach for Exploring New Generative Art
A Practice-based Approach for Exploring New Generative Art Schemes Gary R. Greenfield University of Richmond, Virginia, USA December, 2005 GAP ‘ 05
Outline The ubiquity of GA schemes ¡ Standard search paradigm ¡ Non-interactive genetic algorithms ¡ Our practice-based approach ¡ Conclusions ¡
The ubiquity of GA schemes Biomorphs – Dawkins ’ 89 ¡ Evolving Expressions – Sims ’ 91 ¡ Mutator - Todd & Latham ’ 92 ¡ Fractals – Sprott ’ 96 ¡ Strange Attractors – Krawczyk ’ 03 ¡ Polynomiography – Kalantari ’ 03 ¡ Self-similar Tilings – Priebe ’ 00 ¡
Spiral Tilings – Palmer ’ 05 ¡ Hyperbolic Spirals – Dunham ’ 03 ¡ Bar Grids – Roelofs ’ 04 ¡ Spirolaterals – Krawczyk ’ 00 ¡ Component Sculptures – Hart ’ 03 ¡ Fermat Spirals – Krawczyk ’ 05 ¡ TSP Art – Galanter ’ 04 ¡
TSP Art – Kaplan & Bosch ’ 05 ¡ Polar Transformations – Bleicher ’ 04 ¡ Reaction Diffusion – Behravan & Carlisle ’ 04 ¡ ACO – Aupetit et al ’ 03 ¡ Cellular Morphognesis – Eggenberger ’ 97 ¡
Standard search paradigm ¡ Interactive genetic algorithm IGA - slow and cumbersome - subject to user fatigue - “novelty” generator? (Dorin ’ 01) - aesthetic intent? (Mc. Cormack ’ 05)
Evolving Expressions IGA (Greenfield ’ 92 -’ 96) Modeled after Sims ¡ User interface to control genetics ¡ Features - Node iteration - External image acquisition - Palette management - 2 D and 3 D imaging ¡
Slippage I-III
Eerily Slow
Expressionism IV
Re-coloring Example
A
Non-interactive genetic algorithms ¡ ¡ ¡ Baluja et al ’ 94 Rooke ’ 98 Machado & Cardoso ’ 98 Open Problem : ¡ To derive fitness functions that are capable of measuring human aesthetic properties of phenotypes. (Mc. Cormack ’ 05)
A fitness function taxonomy ¡ ¡ ¡ Positive Feedback - simulated co-evolution - neural nets (Cardoso et al ’ 98) Negative Feedback - simulated immune systems (Romero et al ’ 05) - simulated diseases (Dorin ’ 05) Direct Control - user-designed Indirect Control - multi-objective optimization - ant colony optimization Learning - image analogies (Hertzmann et al ’ 01) - simulated gaze data
Our practice-based approach Co-evolutionary framework (’ 00) ¡ Color segmentation analysis (’ 02) ¡ Multi-objective Optimization (’ 03) ¡ Virtual ant paintings (’ 05) ¡ Cellular morphogenesis (’ 05) ¡ Serial polar transf. motifs (’ 05) ¡
Co-evolutionary framework (’ 00)
Host-parasite mechanics
Fitness calculation
Color segmentation analysis (’ 02)
Images with segmentations
Fitness is responsive to the segmentation geometry
Multi-objective Optimization (’ 03)
Virtual ant paintings (’ 05) ¡ Fitness using arithmetic expressions of exploitation (Nf) and exploration (Nv) measurements
Fitness ~ Nf / Nv
Fitness ~ Nf + Nv
Fitness ~ (Nf) (Nv)
Fitness ~ Nv
Cellular morphogenesis (’ 05)
The Void Series
Serial polar transformation motifs (’ 05)
Randomly generated
Evolved motifs ¡ ¡ Fitness ~ min or max of “boundary” pixel count (genome “length” fixed) Most, average, and least fit for a min run
Another “min” run
A “max” run
Detail
Simulated Robot Paintings ’ 05 ¡ Performance measurements Np - # squares painted Nb - # forward collisions Ns - # couldn’t move Nc - # color sense hits
Fitness ~ weighted linear comb.
Fitness ~ (Nb)(Nc)
Fitness ~ (Ns)(Nc) + (Np)(Nb)
Conclusions Devise appropriate image assessment parameters ¡ Consider fitness function taxonomy ¡ Rely on practice-based implementation ¡
Thank-you! ¡ ggreenfi@richmond. edu ¡ http: //www. mathcs. richmond. edu/~ggreenfi/
- Slides: 56