Reinforcement Learning for Topographic Mappings

We utilise reinforcement learning in the context of unsupervised exploratory data analysis. We first extend a reinforcement learning algorithm which has previously been shown to cluster data. Our first extension involves creating an underlying latent space with some pre-defined structure which enables us to create a topology preserving mapping. We illustrate two different forms of the underlying mapping on artificial data before showing one visualisation of a real data set. We then combine temporal difference learning with the so-called "cross entropy method" and show how the resulting hybrid too can perform unsupervised data analysis.