As part of my fellowship, I am developing tools that allow cognitive scientists to use GANs for their research. GANs are an incredibly powerful approach to learn generative models underlying any kind of data. In practice, we can use GANs to generate (a theoretically infinite number of) experimental stimuli. The main challenge is to control what we generate, so we can generate useful stuff.
The thumbnail of this project shows a face morph between Sabine Schmitz (RIP) and Sarah Connor (not the terminator one). I am currently using GAN-generated face stimuli to study face perception in different human subgroups.
Pardon the compression. The above is a technique inspired by Son et al.’s scene wheels.