A machine learns a landscape

This nearly hour-long video installation unveils an evolving landscape as viewed from a train window, entirely produced by a Generative Adversarial Network (GAN). The piece is characterized by its lack of post-processing or manual editing, with the visuals representing the live output from the algorithm’s continuous learning process. Every 20 seconds, the system updates to enhance the realism and detail of the imagery. The artwork is powered by a GAN, which includes two neural networks: the generator and the discriminator, trained simultaneously. The generator begins by creating images from a random noise distribution, while the discriminator evaluates these images against a dataset of train window videos. The objective is for the generator to produce images so lifelike that the discriminator cannot tell them apart from actual videos. Throughout the duration of the video, the GAN refines its output by learning to predict and generate successive frames from its preceding outputs. Henry has employed this technology to replicate the fleeting and sometimes blurred views typical of train travel. Starting with rudimentary and less detailed visuals, the artwork gradually presents more refined and complex images as the GAN learns from its initial outputs. This progression not only demonstrates the GAN’s capability to detect and replicate complex patterns autonomously but also reflects the transient essence of rail travel. The evolving imagery captures scene dynamics such as the faster movement of the foreground compared to the background, emphasizing the machine's independent learning achievements. Accompanying the visual experience is a custom score curated by music supervisor Legio X with the composition by Carter Mullin, enhancing the immersive experience of the installation.