A Generative adversarial network or (GAN) is a type of artificial intelligence machine learning system that consists of two neural networks in competition with one another, with the goal being to create new data (images, video, photography ect) based on a training set. An artificial neural network works a lot like a biological brain does, but unlike literal synapse connections the network transmits information through neuron like nodes that process information and distribute it to corresponding nodes. This allows for “learning” to take place, without programmed parameters a neural network is able to identify data based on previous information it has encountered.
What a GAN does, is use these neural networks to not only process data but manufacture new unique data sets based on what information it has to process. The two competing networks are known as the generative network and discriminative network. The generative network is responsible for creating data that is statistically as close as possible to the original source material in that it is able to trick the discriminative network. The discriminative network more or less vets datasets only allowing for datasets that are from its perspective identical to the varying source material.
For example if a GAN was trained from a series of photographs of peoples faces the generative network would through a learning series of trial and error create images that would be accepted by the discriminative network. This is why having a larger selection of source material would lead to more passible or realistic final products.
All Ganmojis were created from an open source GAN model made available through the application Runway.ml. From this I fed the model all 3,304 current IOS Emojis as source material. The GAN produced a series of 100 unique icons over 3,000 epochs. The first 102 emojis created are refered to as Gen1, as the model learns through various iterations each subsequent batch is labeled a following generation: Gen2, Gen3 ect...
My name is Ben Glass and I am a visual artist and current undergraduate studying within the Film Video New Media department at the School of the Art Institute Chicago. As part of my on going studies into generated adversarial networks as an artistic medium I created this project in hopes to introduce this emerging technology in relation to art making. I can find no better example of prolific pictoral iconography in modern society than the Emoji. With such negative association with GAN technology, what with the advent of the "DeepFake", I found the Emoji to be a source material that is as recognizable as it is digestible for the majority of viewers.
Runway.ml is an application that allows innovators, artists and creatives to create, test and share artificial intelligence capabilities and projects. The application allows usres to Browse, Train, Experiment and Integrate artificial intelegnce models.