Plenary: Fake Images through Learning of Fast Evolutionary GAN

Generative Adversarial Networks (GAN) are well known for their capability of generating photo-realistic images or data collections that appear real. One of state-of-the-art approaches in GAN is Evolutionary GAN (E-GAN) which can outperform other GAN methods by leveraging the advantages of evolutionary computing, including population based search, mutation and elitism operators. Evolutionary search is often demanding in terms of resources, e.g. computational power and time. That limits its applicability when resource is limited. Partial Transfer learning can improve the efficiency of E-GAN. With the method, the training of generators and discriminators can be achieved with smaller data sets as features can be transferred across different stages of training. Besides the improved performance, it requires less resources to achieve a similar performance as E-GAN. Through partial transfer learning we can speed up generative adversarial learning.