creative design using collaborative interactive genetic algorithms
DESCRIPTION
Dissertation defense. I propose a computational model of creative design based on collaborative interactive genetic algorithms. I test the computational model on two case studies: floorplanning and 3D modeling.TRANSCRIPT
Creative Design Using Collaborative Interactive Genetic Algorithms
Juan C. QuirozPhD Dissertation DefenseThursday April 29, 2010
Department of Computer Science & EngineeringUniversity of Nevada, Reno
Outline
1. Creativity in Design
2. Collaborative Interactive Genetic Algorithms
3. Reducing User Fatigue in Interactive Genetic Algorithms
4. Testing Our Computational Model of Creative Design
5. Contributions
Design
• Conceptual design• Detailed design• Evaluation • Iterative redesign
Conceptual Design
• Initially conceiving and elaborating solutions that meet a set of requirements
• Change in requirements• Subjective evaluation of alternative design
concepts– Aesthetics and other subjective criteria
• Collaboration
Creativity
• Novel and useful• Role of collaboration
Individual Work
Individual
PeersWork
Computational Model of Creative Design
• Allows for subjective exploration of solutions• Supports collaboration• Has the potential to generate creative
solutions
Main Claim
Collaborative interactive genetic algorithms are a viable computational
model of creative design
Outline
1. Creativity in Design
2. Collaborative Interactive Genetic Algorithms
3. Reducing User Fatigue in Interactive Genetic Algorithms
4. Testing Our Computational Model of Creative Design
5. Contributions
Collaborative Interactive Genetic Algorithm
• Population based search technique– Natural selection– Survival of the fittest
CollaborativeInteractive Genetic Algorithms (IGAs)
• Fashion design (Kim 2000)• Micromachine design (Kamalian 2005)• Music, editorial design (Takagi 2001)• Traveling salesman problem (Louis 1999)
Collaborative Interactive Genetic Algorithm
Creative Design
• Floorplans with rectangular rooms• Purposely shifting the focus of
the search space– Circular rooms– Ellipsoid rooms– Star-shaped rooms
Creative Design
Sharing Solutions
IGAP: Interactive Genetic Algorithm Peer to Peer
Outline
1. Creativity in Design
2. Collaborative Interactive Genetic Algorithms
3. Reducing User Fatigue in Interactive Genetic Algorithms
4. Testing Our Computational Model of Creative Design
5. Contributions
User Fatigue in Interactive Genetic Algorithms
• Genetic Algorithms tend to rely on– Large populations– Many generations
• Suboptimal solutions• Noisy fitness
Fitness Interpolation
• Pick the best solution every nth generation
Experimental Setup
• Test on the onemax problem• Subset methods– Best n, best n/2 and worst n/2, random n, PCA n
• Subset size• Gaussian noise• Collaboration
Experimental Setup
• Simulated user input– 20 user evaluations– Greedy user always picks the solution with most
ones• 30 independent runs• Step sizes of 1, 2, 5• Subset size 9
Boxplots
Maximum
Minimum
Median
Upper quartile
Lower quartile
Outlier
Outlier
Subset Methods
Subset Size
Step size 1Step size 2
Step size 5
Subset Size
No Noise vs Gaussian Noise with Sigma=1
Step size 1Step size 2
Step size 5
Noise
Number of Peers
Summary
• Users can effectively bias evolution towards high fitness solutions– Subset size– Noise– Collaboration
Outline
1. Creativity in Design
2. Collaborative Interactive Genetic Algorithms
3. Reducing User Fatigue in Interactive Genetic Algorithms
4. Testing Our Computational Model of Creative Design
5. Contributions
Goals
• User studies– Solutions created individually– Solutions created collaboratively
• Show that solutions created collaboratively are more creative
First User Study: Floorplanning
7 8 9 10 11 12
Bedroom
Living RoomEating area
Bathroom
Collaborative Floorplanning
User’s Individuals Peers’ Individuals
Pilot: Experimental Setup
• Requirements– Design a floorplan for a 2 bedroom, 1 bathroom
apartment– Living room should face north-west– The two bedrooms should not have a common
wall– At least one of the bedrooms should have direct
access to the bathroom
1 2 3 4 5 6
Pilot: Experimental Setup
• Four colleagues and I evolved floorplans– Individually– Collaboratively
• Ten computer science graduate students evaluated the designs by taking a survey
• The plans were evaluated for creative content based on practicality and originality
1 2 3 4 5 6
Floorplan Results
1 2 3 4 5 6
7 8 9 10 11 12
ResultsDESIGN # PRACTICALITY ORIGINALITY RANK
1 2.7 3.1 6 2 2.9 2.7 8 3 2.4 3.3 7 4 4.4 3.3 2 6 4.0 3.2 3 7 2.2 3.4 8 8 2.8 3.4 5 10 3.2 3.5 4 11 4.2 3.6 1 12 1.6 3.8 10
7 8 9 10 11 12
1 2 3 4 5 6
Floorplanning User Study: Experimental Setup
• Requirements– Create a floorplan for a 2 bedroom, 1 bathroom
apartment– Bathrooms close to the bedrooms– Bathrooms far from kitchen and dining areas
Floorplanning User Study: Experimental Setup
• Participants:– 8 women, 12 men
• Five groups of size four• Agenda
1. Tutorial2. Create individual floorplan3. Create collaborative floorplan4. Evaluation of floorplans
Evaluation Criteria1. Appealing – unappealing2. Average – revolutionary3. Commonplace – original4. Conventional –
unconventional5. Dull – exciting6. Fresh - routine7. Novel – predictable8. Unique – ordinary9. Usual - unusual10. Meets all requirements -
does not meet requirements
• Creative Product Semantic Scale
• Seven point Likert scale
Hypothesis
• Is collaboration amongst peers sufficient to allow for the potential to produce creative solutions?
• Designs evolved collaboratively will consistently rank higher in the evaluation criteria.
ResultsEvaluation Criterion Desired Ind. Avg. Coll. Avg. P-value
Appealing - Unappealing Low 4.08 4.39 0.439
Average - Revolutionary High 3.76 4.34 0.047Commonplace - Original High 3.97 4.68 0.021
Conventional - Unconventional
High 4.03 4.41 0.355
Dull – Exciting High 3.65 3.93 0.326
Fresh – Routine Low 3.82 3.68 0.810
Novel - Predictable Low 3.55 3.40 0.697Unique - Ordinary Low 3.49 3.11 0.251Usual - Unusual High 4.21 4.51 0.395
Meets All Req. -Does Not Meet Req.
Low 2.63 2.83 0.779
Discussion
• Ambiguity in evaluation criteria– Appealing – unappealing– Positive – Negative (?)– Negative – Positive (?)
• Applicability of evaluation criteria– “Exciting”– Domain expert vs. student
• Participants created only 1 collaborative floorplan and 1 individual floorplan
• Simple graphic representation
Second User Study: 3D Modeling
• Vertex programs– p.x += 20– p.x += sin(time)
Sample Ninja Transformations
Collaborative Setup
• User 1– Equations that modify the
x and z coordinates
• User 2– Equations that modify the
y and z coordinates
• After collaboration– Equations that modify the
x, y, and z coordinates
Experimental Setup
• Design Phase– Two groups of 10 participants
• Evaluation Phase– On-site evaluation• 20 participants
– Online evaluation• 16 participants
Experimental Setup
• Groups of 2• Agenda
1. Tutorial2. Creating 3D models3. Picking solutions for
the evaluation phase
Individually
Collaboratively
Individually
Collaboratively
Individually
Collaboratively
Design Phase
Female
Ninja
Robot
Robot
Ninja
Female
Individually
Collaboratively
Individually
Collaboratively
Individually
Collaboratively
Female
Ninja
Robot
Female
Ninja
Robot
Evaluation Phase
• 7 point Likert scale• Creative Product Semantic Scale• The transformation is:
– Extremely creative – Not Creative At All
• The transformation can be used in a video game.
• The transformation with minor tweaks can be used in a video game.
• The transformation is novel.• The transformation is surprising.
Individually
Collaboratively
Individually
Collaboratively
Individually
Collaboratively
Results
Individual Collaborative
The transformation is creative.
The transformation can be used in a video game.
The transformation with minor tweaks can be used in a video game.
The transformation is novel.
The transformation is surprising.
Online Evaluation
• Best individually created models• Best collaboratively created models• Evaluation Criteria– The transformation is creative.– The transformation can be used in a video game.– The transformation is novel.– The transformation is surprising.– Which of the two rows did you like the most?– Which of the two rows is the most creative?
Results
• Individually created models vs collaboratively created models– No statistically significant results
Results
• Which of the two rows did you like the most?– 8 participants picked the
individual row– 7 participants picked the
collaborative row– 1 participant did not answer
• Which of the two rows is the most creative?– 3 participants picked the
individual row– 13 participants picked the
collaborative row
IndividualCollaborativeNA
IndividualCollaborative
Discussion
• Different 3D models• Lack of context• Online Evaluation Nuances– Switching windows– 15 second average– Rewinding– Scoring the row of individually created models first– Indecisive participants and median scores
Outline
1. Creativity in Design
2. Collaborative Interactive Genetic Algorithms
3. Reducing User Fatigue in Interactive Genetic Algorithms
4. Testing Our Computational Model of Creative Design
5. Contributions
Contributions
• A new computational model of creative design– Subjective exploration of solutions– Integrates collaboration
• Implementation of IGAP framework: Interactive Genetic Algorithm Peer to Peer
• Analysis of our fitness interpolation technique in the onemax problem
Contributions
• Floorplanning pilot– Collaborative solutions were considered more
original• Floorplanning user study– Collaborative solutions were considered more
original and revolutionary• 3D Modeling user study– 13 out of 16 participants picked row of
collaborative of solutions as the most creative
Future Work
• Conduct additional user studies– Long term user studies with design teams
• Refine and test IGAP framework• Machine learning
Acknowledgments
• Dr. Sushil Louis• Dr. Bobby Bryant• Dr. Swatee Naik• Dr. Sergiu Dascalu• Dr. Amit Banerjee• Dr. Darren Platt• Study participants
– Students, adult volunteers, and faculty• This work was supported in part by contract number N00014-
05-1-0709 from the Office of Naval Research and the National Science Foundation under Grant no. 0447416.
Questions?
Thank you!
www.cse.unr.edu/~quiroz