Our paper (joint with Amelie Levray) on learning credal sum-product networks has been accepted to AKBC. Such networks, along with other types of probabilistic circuits, are attractive because they guarantee that certain types of probability estimation queries can be computed in time linear in the size of the network. The problem we tackle is how the learning should be defined when there is missing or incomplete data, leading to an account based on imprecise probabilities. Preprint here.