Embedded Files
Ian Goodfellow (1985-) is a machine learning researcher, previously working for Google, mostly known for the invention of Generative Adversarial Networks (GANs) [1]. While he is not an old new media artist, I think he has made a very notable contribution to the field of media art. Through his invention of GANs, he has basically started a new field, using neural networks to generative many different kinds of visual images. Since the first publication of 2014, this field has basically exploded (figure 1).
While other techniques were used before the invention of GANs to generate new images, the core idea that Ian Goodfellow came up with was to create a kind of competition between two neural networks: a Generator, which tries to create images which look real, and a Discriminator, which tries to detect false images (figure 2). These two neural networks are basically in a game, competing with each other, and over time they both improve and thus a GAN is created which can create images which look impressively realistic.
Some well known examples include This Person Does Not Exist [2], a website which outputs images generated by a GAN of faces; CycleGAN [3], which can change the style of images to the style of another set of images, e.g. change a painting into the style of Monet (figure 3); and Portrait of Edmond Belamy [4], a painting generated using a GAN.
While long discussions can be had on the implications of using AI in this field, as well as whether this is actually art, it cannot be denied that the invention of GANs has ignited new creations and will certainly have a large impact on the future of media art.
Figure 1: Number of publications related to GANs per month
Figure 2: Workings of a GAN
Figure 3: Examples of CycleGAN output
Page updated
Google Sites
Report abuse