Latent Space

2017, digital video without audio, 45 minute loop
2021 iteration, digital video with audio, 6 hour loop

In artificial intelligence ‘Latent Space’ refers to a mathematical space which maps what a neural network has learnt from training images. Once it has been trained it understands all images of trees as existing in a specific area, and all images of birds in another.

The neural network can be reverse engineered to create fake images from these coordinates₁. But what if it was given a new path to travel between these recognisable points, instead moving through the in-between space?

Latent Space is a video snap-shot of an A.I. algorithm in its infancy trained using 14.2 million photographs₂ continuously producing new images.

1) Plug & Play Generative Networks: Conditional Iterative Generation of Images in Latent Space (2016)
2) ImageNet: A Large-Scale Hierarchical Image Database (2009)
Special thanks to Anh Nguyen et al. at Evolving-AI for their research.