Truncation #1: Daubing the Latent Space

This exclusive work is part of the "Truncation" series by Liam Power. What do neural networks look like? How does a computer learn? Through a dataset trained on the cumulative history of artworks from antiquity to modernism, a neural network steps through an interpolated series of numbers. These numbers by chance correlate to an artwork, a hyperreal simulation of the original. In this work, we get a glimpse inside the neural processing unit of a computer. Tensors shift and flow. A generative network splits attention into hemispheres. Artificial paint shifts in smooth interpolated waves, truncation incrementing in the latent space. Simulated neurons fire infinite variations on a machine dream. A simulated soundscape traverses the network in a feedback loop. Includes a HQ .mp4 for inclusion in offline/offline realities.