In his most recent research paper he presents sketch-rnn, a recurrent neural network (RNN) able to construct stroke-based drawings of common objects. The model is trained on thousands of crude human-drawn images representing hundreds of classes. They outline a framework for conditional and unconditional sketch generation, and describe new robust training methods for generating coherent sketch drawings in a vector format.
In this talk I will discuss some of my experience with getting neural networks to do interesting things, without clear useful goals in mind. For example, I will show how we can make simple dataset interesting by getting a neural network to enlarge images without being explicitly being trained to do so. I will also discuss the use of neural networks to generate vector sketches, and finally, to generate entire worlds.
THURSDAY, JUNE 7th • 10:30AM • McGUIRE THEATER