I built a DC-GAN based neural network using Tensorflow/Keras and trained it on 28x28 pixel images of airplanes, after which it generated the above airplane “photographs.” A current work-in-development called Synthetic Prairie will use this DC-GAN along with Wavegan audio-generating networks to create an artificial landscape installation.

I created a Wavegan-structured neural network using Tensorflow/Keras and trained and optimized it for generating single, spoken words. A current installation-in-development called Synthetic Prairie will use Wavegan and other audio-generating CNNs to create an artifical prairie soundscape; the networks will likely have to be adapted for longer-form audio that emulates a natural soundscape/field recording.

A Twitterbot that tweets angsty moth poetry. Based on the Tracery library by Dr. Kate Compton.

Algorithmic granulation instrument in development in Python; soon to switch to SuperCollider. This is to be a replacement for the Ableton Live + Max4Live setup I’ve used since 2014.

A series of incomplete drawings on graph paper were created by filling in a selection of grid spaces with colorful shapes. The positions within the grid, shapes, and colors were turned into training datasets for a different random forest model per drawing; the models then predicted how each drawing should be completed.