The project attempts to explore the processes of machine learning and neural networks through their unconventional and somewhat counterintuitive use. It depicts the process of training a neural network all the way from collecting the initial images to animating its final products. The principle that the project explores is related to the processes of training in computer vision, where neural networks are exposed to thousands of images of specific objects, and the “algorithmic summary” created along the way changes with every image and gradually approaches some kind of a visual essence of individual objects. What is specific about this project is that the initial database used is not homogenous, but the product of a generative process. It is therefore first and foremost an experiment – we don’t know how the neural network will interpret the distinctly diverse bank of images, and we’re interested in how the final product will reflect the nature of the process.
Although the broad collection of images that form the initial visual bank, displayed on the first screen, seems completely random, it is loosely based on the author’s visual interests and therefore forms a unique moodboard. The work can therefore be understood as an anticipation of future creative processes enriched with neural networks, where artists equipped with such tools will be able to create new content through associative combinatorics alone.
Photogallery of the exhibition (photo: Andrej Lamut)