Recently I’ve been playing with Generative Adversarial Networks, which are a new flavour of neural network architecture which can be used to generate new representations from a corpus of data. That data can be images, speech, music, pretty much anything that can be represented as data.
In this case, I’m using a GAN to generate new images from a set of pictures.
I fed my network the mugshots of all current UK MPs, in an attempt to find what the politicians of latent space might look like.
Technically speaking, 649 images is way too few to get any decent generalisation. But for the purposes of art, the results are perfect.
Herewith, 32 algorithmically generated politicians:
