Intro:
In 2014, Ian Goodfellow and his colleagues proposed the machine learning architecture of the Generative Adversarial Network (GAN), a generative model that can produce images, text, or music. [1] In 2016, Tim Salimans alongside Goodfellow and other OpenAI researchers responded to the initial GAN paper with a paper on improved techniques for training GANs that discussed methods to stabilize training and encourage convergence. [2] This research directly led to the development of the Wasserstein GAN (WGAN) and subsequent papers by Martin Arjovsky and his colleagues. [3][4] The Wasserstein GAN aimed to reduce mode collapse and create more stable training than the original GAN. Today, there are many interesting applications of Wasserstein GANs including data augmentation for emotion recognition [5] and singing voice synthesizers [6].