A temporal approach to belief revision in knowledge bases

A new technique for the efficient revision of beliefs in knowledge bases for real-time applications is presented. The technique is based on the notion of symbolic transaction times and has the advantage of providing a temporal history of implicit as well as explicit beliefs, essential for diagnosis and optimization. It combines the efficiency of justification-based truth maintenance systems (TMSs) with the multiple-context handling capability of assumption-based TMSs. However, it involves neither dependency-directed backtracking nor label update propagation algorithms. Hence both forward and backward belief revision is achieved more efficiently. The technique may easily be adapted to reflect various belief revision semantics by embedding these notions into a constraint solver subsystem. An instance of the proposed technique has been used in the development of an air traffic management system