With the HDP knowledge base there are thousands of paths that lead to a node such as pulmonary congestion. Most of these paths are never used, but without further information they must be investigated. In the version of the HDP without temporal reasoning, we pre-computed all of the paths to provide a fast way of picking the most likely candidates for hypotheses and picking the best causal paths. Temporal reasoning provides a way to extent this mechanism. Even though the times of nodes are not known before a case is entered, the minimum and maximum extent are known from the temporal properties in the knowledge base. These can be transferred down the causal chain by applying each new constraint to the one above it. When there is an inconsistency, that causal path can be eliminated.
Therefore, we added properties min-exist and max-exist to each node in the computed causal path. The node constraint min-exist is determined by the onset and the max-exist is determined by the max-exist and persist properties. In the causal path, if the max-exist of the cause is less than the delay or onset or if the min-exist of the cause is greater than the max-exist of the effect, the causal path is stopped. In this way we were able to eliminate about 20%of the causal pathways that were generated in the older version of the HDP.
This also allows the computation of time bounds on nodes before all of the causality has been determined. In the example case, it is possible to determine that the earliest possible begin time for PC is two weeks, because that is the earliest time for any causal path (in the very limited model) that causes PC. This in turn assures that any hypothesis that would require PC longer than two weeks would be eliminated without further computation. Thus, pre-computing the implications of the temporal constraints allows the program to make optimal use of them.