Probabilistic Databases under Updates: Boolean Query Evaluation and Ranked Enumeration

We consider tuple-independent probabilistic databases in a dynamic setting, where tuples can be inserted or deleted. In this context we are interested in efficient data structures for maintaining the query result of Boolean as well as non-Boolean queries. For Boolean queries, we show how the known lifted inference rules can be made dynamic, so that they support single-tuple updates with only a constant number of arithmetic operations. As a consequence, we obtain that the probability of every safe UCQ can be maintained with constant update time. For non-Boolean queries, our task is to enumerate all result tuples ranked by their probability. We develop lifted inference rules for non-Boolean queries, and, based on these rules, provide a dynamic data structure that allows both log-time updates and ranked enumeration with logarithmic delay. As an application, we identify a fragment of non-repeating conjunctive queries that supports log-time updates as well as log-delay ranked enumeration. This characterisation is tight under the OMv-conjecture. This is joint work with Christoph Berkholz. The presentation is based on our paper at PODS 2021.

Authors

• Maximilian Merz

Keywords

No keywords are indexed for this paper.