Combining the lazy label evaluation with focusing techniques in an ATMS
Mugur Tătar · European Conference on Artificial Intelligence · 1994
For large problems the ATMS often becomes the main resource consumer in any reasoning system. We propose an archi- tecture (the 2vATMS) that combines the advantages of two tech- niques aiming to reduce the complexity of the ATMS tasks: (i) fo- cusing on a few possible worlds, which abandons the requirement of global label completeness; and (ii) the lazy label evaluation, which avoids the label update work as long as there is no evi- dence that it would be relevant for the problem solver. The 2vATMS integrates two logically dependent views on data: The first view stores by which focus environments a node is sup- ported. This information is easy to maintain and corresponds to the information provided by a set of simple monotonic TMSs. The second view provides more detailed information, i.e. about the dependence on the assumptions. The labels of the detailed view are similar to the ones computed by a focused ATMS, but their maintenance is ensured only by request. The content of the first view helps to achieve a very tight control of the environment propagation in the detailed view. Moreover, the problem solver has the opportunity to decide when the information offered by the monotonic TMSs is sufficient, and when the switch to the more detailed view is valuable.