Learning for semideterministic reasoning
Chengjiang Mao, Daniel L. Chester · 1999
Traditional reasoning strategy, which is considered as non-deterministic reasoning in this dissertation, has dominated AI since the birth of this field. Although it makes it much easier for programmers to write typical rule-based programs to solve AI problems, its poor efficiency, compared with conventional programming, is still an open problem. The semi-deterministic reasoning which is accomplished here works toward overcoming this problem. Semi-deterministic reasoning consists of two stages: deterministic reasoning to select relevant rules for an input problem instance and non-deterministic reasoning to infer results based on the selected relevant rules. It improves reasoning efficiency because the costly conventional non-deterministic reasoning at the second stage is performed in a much reduced set of rules from the original knowledge base. Semi-deterministic reasoning is based on a new knowledge base organization, which is transferred from the common knowledge base organizations that support non-deterministic reasoning. This transformation is automatically realized by an integrated learning system called Thought-Prolog and Thought-Prolog/C, which are described in this dissertation. Thought-Prolog and Thought-Prolog/C are specifically designed to reorganize a typical unorganized Prolog program to support semi-deterministic reasoning.