Focusing the ATMS

Many problems having enormous search spaces can nevertheless be solved because only a very small fraction of that space need be traversed to find the needed solution(s). The ability of assumption-based truth maintenance systems to rapidly switch and compare contexts is an advantage for such problems, but existing ATMS techniques generally perform badly on them. This paper describes a new strategy, the implied-by strategy, for using the ATMS that efficiently supports problem solving in domains which are infinite and where the inference engine must retain tight control in the course of problem solving. We describe the three mechanisms required to implement this strategy: focus environments, implied-by consumers, and contradiction conaumera. We compare the implied-by strategy to previous ATMS strategies, and demonstrate its effectiveness by performance comparisons in two simple domains. Finally, we discuss the implications for parallel processing and future ATMS design.