Title: Intelligent Caching in Hybrid Knowledge Bases
Authors: Sibel Adali, V.S. Subrahmanian.
Abstract
Lu, Nerode and Subrahmanian have developed the concept of a hybrid knowledge base (HKB). HKBs may be used to access and logically integrate heterogeneous information from external databases and/or software packages using functions already defined in those domains. When executing queries to HKBs, we need to execute these externally defined functions, and this is often an expensive process, especially when these external programs are located at remote sites on, say a network. Consequently, the problem of minimizing access to such external routines without compromising completeness is an important one. In this paper, we will present three procedures that make use of caches with successively larger degrees of ``intelligence.'' The first method uses invariants -- information about the relationship between various external calls -- to cut down on the number of external calls. The second method shows that the use of caches and invariants may be itself encoded as an external program that can then be executed using the ordinary method. The third method builds on the second by showing that in various cases, invariants may be semantically composed together, to further reduce the number of external calls.