The BMO Lab is proud to announce that the research paper produced by BMO Lab summer intern Mengyu Yang in conjunction with Creative Technologist in Residence Xavier Snelgrove and Lab Director David Rokeby has been accepted for the Workshop on Machine Learning for Creativity and Design at NeurIPS 2020, one of the most important international Machine Learning conferences.
The paper represents the first steps in a larger research program which is looking into ways that enable live performers to directly engage with the ‘latent space’ of trained machine learning networks. This particular paper looks at a neural network called StyleGAN2 which generates very realistic faces of non-existent people. It describes a technique by which the machine can learn ways of controlling very specific aspects of the generated face (i.e. changing the various possible positions of the mouth or eyes) without affecting other aspects.
This paper is a direct offshoot of the early experiment presented at the official opening of the BMO Lab where a performer used various aspects of her physical pose to control aspects of a generated face in real-time.