Tomita was a neural network hack I experimented with while working at Kano. Using 10,000 user created artworks made in Kano's Make Art software which were fed into a neural network that learnt how to create new artworks from the example source code.
Make Art is a code based drawing environment which lends itself well to experiments like this because one artwork can flow into the next influencing it in subtle ways.
One interesting part of the data set is the prevalence of the default tutorial code as most Make Art users are just learning to code the most popular creations are variations on the tutorial. It's fascinating how the neural net actually picked up on this and often hallucinates these tutorial creations.