Learning Stylistic Desires and Generating Preferred Designs of Consumers Using Neural Networks and Genetic Algorithms

Consumers have different ideas of what makes a design stylish. Some consumers may want a sporty looking car, while others may want a rugged looking or a fuel-efficient looking car. Can computers learn what it means to satisfy those style-based goals and use this knowledge to generate designs that target style-based goals in design? An experiment was conducted where participants were asked to rate computer generated car profiles for sportiness, ruggedness, beauty, and fuel efficiency. This survey data is used as an indicator of consumer stylistic form preferences, and was used to train Artificial Neural Networks (ANN) for each of the four rating categories. The resulting ANNs were then inverted using a Genetic Algorithm (GA) in order to generate new designs that elicit targeted style goals from consumers.