Breaking the Mould: An Evolutionary Quest for Innovation Through Style Change
Authors
Abstract
An autonomous creative system able to learn, create and innovate is presented.Following previous work on the same topic, the approach explores the interplay
between a classifier and an evolutionary system. The classifier is trained
with famous paintings and images created by the system, learning to distinguish between
these two categories. It is then used to assign fitness, leading to the discovery
of imagery that deviates from that previously created by the system. Additionally,
by taking phenotype similarity into account, we promote the discovery of diverse
images during the course of the evolutionary runs. The images created throughout
the evolutionary runs are added to the training set and the process is repeated. This
iterative process, which includes retraining the classifier, sets the system into a permanent
quest for novelty and innovation. The experimental results obtained across
several iterations are presented and analysed, showing the ability of the system to
consistently produce novel imagery and to identify atypical images.