While the majority of well-known GANs are used for image creations, RadialGAN is designed for numerical analysis. If we consider an example where we want to evaluate how effective a new medical treatment is, we would need to combine data from a number of different hospitals in order to ensure we have enough data to make concrete conclusions. However, this poses problems such as different hospitals measuring outcomes in different ways, using laboratories that give different results in different environments, and so on. In order to address this problem, RadialGAN firstly transforms the dataset from each hospital into latent space, which allows us to hold the data from different sources in a uniform format. From here, the latent space data can be converted into the feature space of each unique dataset.
Each dataset considered by the RadialGAN has an encoder neural network...
