Budget Amount *help |
¥12,180,000 (Direct Cost: ¥11,100,000、Indirect Cost: ¥1,080,000)
Fiscal Year 2007: ¥4,680,000 (Direct Cost: ¥3,600,000、Indirect Cost: ¥1,080,000)
Fiscal Year 2006: ¥3,300,000 (Direct Cost: ¥3,300,000)
Fiscal Year 2005: ¥4,200,000 (Direct Cost: ¥4,200,000)
|
Research Abstract |
In this research, we developed new advanced methods for hypothesis finding and implemented efficient hypothesis-finding systems based on consequence-finding techniques. The proposed systems, SOLAR and CF-induction, compute abductive and inductive hypotheses, respectively. Research results can be summarized as the following three items. 1. Basic theories for advanced hypothesis-finding (Iwanuma, Inoue, Nabeshima). We developed theoretical results on equivalence of abductive theories, generality of default theories and logic programs, and induction of causal laws in action descriptions. We also studied some aspects of circumscriptive induction, which is a new integrated framework of explanatory induction and descriptive induction, and proposed a new first-order approximation method, called pointwise circumscriptive induction, which enables us to derive some interesting hypotheses by the ordinary consequence finding calculus. 2. Development of an efficient consequence-finding system (Nabeshi
… More
ma, Iwanuma, Inoue). We showed that a size-preserving upside-down transformation of any SOL/Connection tableau can be achieved by the folding-up operation, and gave a size-optimal consequence finding calculus for an incrementally axiom-increasing environment. We also developed a consequence-finding method with non-stable production fields and applied it to complete abduction from full clausal theories. We further implemented a new version of the consequence-finding system SOLAR with various state-of-the-art pruning techniques. Handling axioms with equality is also investigated in the SOLAR framework, and a prototype system of a C++ version of SOLAR is firstly realized. 3. Hypothesis-finding by CF-induction and its application (Inoue). CF-induction is a sound and complete inductive logic programming system to compute hypotheses from full-clausal theories.CF-induction consists of several nondeterministic procedures, and in particular its generalization procedure is assumed to be any complete combination of inductive operators. We thus proposed a method to reduce possible combinations of generalization operators by preserving the soundness and completeness of CF-induction. Next, we applied CF-induction to estimation of possible enzymatic reaction states in metabolic pathways. In this work, we showed that CF-induction can compute not only possible states of enzymatic reactions but also causal relations that are missing in the current background theory. Less
|