Creative Design Using Collaborative Interactive Genetic Algorithms

I propose a computational model of creative design based on collaborative interactive genetic algorithms. My model enhances creativity in the conceptual design phase by allowing designers to guide genetic algorithms in order to breed new design ideas quickly, and by supporting team collaboration through the sharing of solutions among designers. Finally, I attack the problem of user fatigue in using an interactive genetic algorithm to evolve design solutions.First, I present a computational model of creative design based on collaborative interactive genetic algorithms. I test my model on the problems of floorplanning and the generation of transformations for 3D models. I support collaboration by allowing individual designers to view each others' designs during the evolutionary process and to share designs via case injection. Collaboration causes changes in the design space, introducing the potential to generate creative solutions. Results show that solutions generated with collaborative interactive genetic algorithms are more creative than solutions generated without collaboration.Second, an interactive genetic algorithm (IGA) which requires the user to provide a large number of user evaluations for many generations can lead to user fatigue. Reducing user fatigue is a critical challenge in my research since my computational model relies on IGAs. Fitness interpolation allows for the fitness values of individuals in an IGA to be estimated from a representative set of individuals. Such a technique effectively reduces the number of evaluations needed from a user, since the user is only required to evaluate the representative set of individuals. I present a fitness interpolation technique consisting of picking the solution the user likes best, and estimating the fitness of every other individual in the population based on similarity to the user selected best. In addition, the user is asked to make the choice once every t generations. I apply my technique to the well understood Onemax problem allowing us to empirically compare the performance of my proposed technique against a standard IGA. Results show that I can reduce the number of user evaluations performed by a user by an order of magnitude compared to a standard IGA. By using my fitness interpolation technique a user can effectively bias the IGA search towards solutions of interest