" Generative Adversarial Networks (GANs) were proposed in 2014 as a new method efficiently producing realistic images. Since their original formulation, GANs have been successfully applied to different domains of machine learning: video, sound generation, and image editing. However, our theoretical understanding of GANs remains limited.
In this presentation, we will first define the overall framework of GANs and illustrate their main applications. Then, we will focus on a cousin approach called Wasserstein GANs (WGANs). This formulation based on the well-known Wasserstein distance has been validated by many empirical studies and brings stabilization in the training process. Finally, motivated by the important question of characterizing the geometrical properties of WGANs, we will show that for a fixed sample size, optimality for WGANs is closely linked with connected paths minimizing the sum of the squared Euclidean distances between the sample points. "