Neural networks (and deep learning in general) have an unlimited potential for generative applications. One of the most well-known out there is DeepDream, which is a sort of reversed computer vision that “sees” images it’s trained on.

But one of the most exciting applications for neural networks is in real-time generation, like music and animation. Since neural networks are trained ahead of time, they can act quickly in practice.

In this video, Philippe Pasquier shows and discusses his work training a neural network on motion-captured emotion, without the use of facial expressions.

This video comes from the Program Generative Art and Computational Creativity.

Generative Art and Computational Creativity

Simon Fraser University