Evolutionary Latent Space Exploration of Generative Adversarial Networks.

Generative Adversarial Networkss (GANs) have gained popularity over the years, presenting state-of-the-art results in the generation of samples that follow the distribution of the input training dataset. While research is being done to make GANs more reliable and able to generate better samples, the exploration of its latent space is not given as much attention. The latent space is unique for each model and is, ultimately, what determines the output from the generator. Usually, a random sample vector is taken from the latent space without regard to which output it produces through the generator. In this paper, we move towards an approach for the generation of latent vectors and traversing the latent space with pre-determined criteria, using different approaches. We focus on the generation of sets of diverse examples by searching in the latent space using Genetic Algorithms and Map Elites. A set of experiments are performed and analysed, comparing the implemented approaches with the traditional approach.