SIGMA (2019) is a duet between my choreographic brain and a variational autoencoder trained on my own improvisational dancing. Each movement is generated from a 256-dimensional latent space that encodes 5-second sequences as single points. I traverse this space in various directions to understand my typical modes of moving and to challenge myself to venture far away from these ideas. Sigma, the standard deviation from my most typical movement, can increase without bound, extending my choreographic mind and body along with it.
All movements in this video are generated from the VAE my colleagues and I trained on my movement data using a Vicon motion-capture studio. A detailed technical description may be found here: https://arxiv.org/abs/1907.05297.
