Variations on Computational Association

Bora Aydintug

A showcase of machine associations based on my artwork.



As a visual artist, I try to evoke curiosity of the audience in regards to their own inner universe and the common universe that we share. I hope to encourage the audience to let their mind wonder and make free associations. When we look at abstract shapes, random or natural patterns, we liken them to other things. Usually, when we are making free associations, we simply liken shapes to other shapes, if we, however, let our minds wonder and remain with the free-associative activity, we will soon start making associations that are increasingly our own. These personal associations are unique in their relationship to our world of meanings, and usually point towards memories of the past. Free associations, in this sense, are the emergence of connections to experiences of the past.

My fluid dynamic visuals are particularly inviting in this regard. I, therefore, wanted to experiment with BigbiGAN, to see what kind of associations this generative model would make, based on these visuals. Upon analyzing the features of an image, BigbiGAN makes a guess as to what it’s seeing and then generates the image of what it assumes to be seeing. By design, the network makes use of its experience of having been shown similar images. Although devoid of consciousness and therefore meaning, the network is sharing its experiences of the past with us.

The results were obtained after many experiments in Runway, which incorporates an implementation of BigbiGAN.


Open Source Cinema, The Neural Aesthetic