Ian Goodfellow Introduces GANs

In one of this story’s most relevant moments, Ian Goodfellow and his colleagues at the University of Montreal introduced the world to Generative Adversarial Networks (GANs), revolutionizing—and I don’t use the word lightly—the field of machine learning. GANs consist of two models: a generative model that captures the data distribution, and a discriminative model that estimates the likelihood of a sample being from the training data or the generative model. This setup creates a minimax game, driving both models to improve until the generative model produces indistinguishable data from actual data. GANs are so significant because they allowed artists to realize the dreams of the early cybernetic theorists and artists who sought to produce art autonomously from the learning and feedback of early machines. This thinking extends to artists such as Victor Vasarely, François Morellet and thinkers like Max Bense. GANs improved upon the genetic algorithms of the ’90s and ’00s that had issues scaling with larger data sets. GANs allowed for the neural network training time to greatly improve. Ultimately, GANs allowed for artists to better generate new aesthetic output that better mimicked the related training data.