Title: Markov properties of determinantal point processes
Authors: Kayvan Sadeghi - University of Cambridge (United Kingdom) [presenting]
Abstract: Determinantal point processes (DPPs) have been widely used in machine learning for statistical modelling. We discuss the conditional independence structure of DPPs. In particular, we show that the induced independence model by DPPs can be naturally captured by bidirected graphs. In addition, we show that the context-specific induced independence models by DPPs (conditioning on variables being all equal to 1) act in the same way as the independence model induced by Gaussian distribution. This leads to context-specific DPP undirected as well as directed acyclic graphical models.