Processing of Inference Queries in Probabilistic Databases

Real world applications like sensor network monitoring, that deal with wide existence of uncertain factors, employ relational databases to describe the probability distribution of all variables in its environment. Such probability distribution data finds extensive usage through inference queries. In practice, however, rather than a single probabilistic inference query, applications pose multiple but usually similar probabilistic interference queries to the system. An environment that involves frequent inference queries on relational databases provides a possibility of applying 'Computation sharing' logic among different queries. CTP is introduced in databases for probabilistic inference queries. Such an approach provides two opportunities for computation sharing. First opportunity is an existence of many common variables that needed to be eliminated during the evaluation of different queries. Second opportunity refers to the variables appearing frequently in the queries that can be cached and reused in later queries. The materialized views are used to cache the intermediate results of the previous infcrence queries, which might be shared with the following queries, and consequently reduce the time cost.