Copulas in Machine Learning (CopML)

The modeling of dependency structures and probability distributions plays a central role in data analysis, especially in machine learning. Copulas are an excellent tool from classical statistics. These allow probability distributions to be decomposed in a clever way so that the modeling is simplified overall, which is also known as the copula trick. Moreover, since copulas are themselves probability distributions, they can be used to simulate new data. Furthermore, they are suitable for application to high-dimensional data. Due to these properties, copulas are already receiving growing attention in the field of machine learning. The goal of the research project Copulas in Machine Learning (CopML) is to further link copulas and machine learning methods. On the one hand, we want to investigate the potential applications of copulas in machine learning. An example of this is the exploration of possible applications in image or language processing. On the other hand, machine learning methods are to be explored which enable the modeling of copulas themselves.