The increasing diversity of scientific and engineering data has driven the development of flexible techniques for inferring probability distributions without assuming a specific parametric family.