Research on Knowledge Discovery based on Consequence Finding

在这项研究中，我们开发了新的不完全知识下的知识发现方法。所提出的系统基于结果发现过程来计算溯因假设和归纳假设。研究成果可概括为以下三个方面。结果发现的高效计算我们采用井上的SOL分解作为结果发现的过程，使其更加高效。特别是，我们开发了SOL-S(Г)表，用于多智能体系统中高效的投机计算和缺省推理。此外，我们用Java重新实现了SOL Tableaux，并开发了一个更快的结果发现过程SOLAR(SOL For Advanced Reason)。结果发现和知识发现的基础理论我们证明了一阶子句理论中答案抽取的SOL归结是完备的。这是对连接表格式中答案完整性的一个公开问题的解决方案。我们还考虑了一个基于结果发现的假设发现过程(称为CF-归纳)，并在CF-归纳中找到了一个完整的泛化方法。此外，我们还建立了说明性归纳和描述性归纳相结合的统一归纳论。该归纳形式化是基于限定的，并使用SOL归结和CF-归纳法计算假设。我们将SOL分解应用于生物信息学，并考虑了扩展外展的使用，这使我们能够去掉和添加假说。

In this research, we developed new methods of knowledge discovery under incomplete knowledge. The proposed systems compute abductive and inductive hypotheses based on consequence-finding procedures. Research results can be summarized as the following three items.1. Efficient computation for consequence findingWe adopted SOL resolution by Inoue as a consequence-finding procedure and made it more efficient. In particular, we developed SOL-S(Г) tableaux for efficient speculative computation in multi-agent systems and default reasoning. Moreover, we re-implemented SOL tableaux in Java, and developed a faster consequence-finding procedure SOLAR (SOL for Advanced Reasoning).2. Basic theories for consequence finding and knowledge discoveryWe proved that SOL resolution is complete for answer extraction in first-order clausal theories. This is a solution of an open problem for answer completeness in a connection tableaux format. We also considered a hypothesis-finding procedure based on consequence finding (called CF-induction), and found a complete method for generalization in CF-induction. Moreover, we established a unified theory for induction, which combines explanatory induction and descriptive induction. This inductive formalization is based on circumscription, and uses SOL resolution and CF-induction for computing hypotheses.3. Evaluation and applications of hvpothesis-finding proceduresWe applied SOL resolution to bioinformatics, and considers the use of extended abduction, which enables us to remove hypotheses as well as addition of them.