(un)stable equilibrium (2019) by Terence Broad is a collection of abstract works generated using neural networks without any training data. Inspired by the generative adversarial networks (GAN) framework, Broad developed an approach made up of two generative neural networks producing images to imitate each other, whilst competing to have more colour diversity. The title of the piece reflects the experimental process of finding a balance of randomness and stability in the training process.
With their blocks of colour and aesthetically pleasing colour combinations, the works in the collection remind us of Mark Rothko’s early abstract paintings and the colour field movement. Yet no images were fed into Broad’s model, much less Rothko’s. Colour is not a data source here, but rather a means of communication: a form of a meditative dialogue between two neural networks, through two artificial representations of colour fields.
The 600 works in the collection are latent space interpolations based on six experiments from the (un)stable equilibrium (2019) series, all done with different variations of loss functions and training parameters. There are 100 colour variations of each experiment. Visually, each experiment is differentiated by spatial composition, proportion and the occurrence of rectangular or circular shapes.