Abstract
Identifying causal relations is crucial for a variety of downstream tasks. Inadditional to observational data, background knowledge (BK), which could beattained from human expertise or experiments, is usually introduced foruncovering causal relations. This raises an open problem that in the presenceof latent variables, what causal relations are identifiable from observationaldata and BK. In this paper, we propose two novel rules for incorporating BK,which offer a new perspective to the open problem. In addition, we show thatthese rules are applicable in some typical causality tasks, such as determiningthe set of possible causal effects with observational data. Our rule-basedapproach enhances the state-of-the-art method by circumventing a process ofenumerating block sets that would otherwise take exponential complexity.