段階的前向き仮説推論システム
スポンサーリンク
概要
- 論文の詳細を見る
Hypothesis-based reasoning is a basic technique for building AI systems based on incomplete knowledge. The hypothesis-based reasoning system derives conclusions despite the absence of total domain knowledge. When we modify the knowledge base, the beliefs of the conclusions should be partially revised on the system. An assumption-based truth maintenance system (ATMS) is a hypothesis-based reasoning system whose inputs are a set of propositional Horn clauses and a set of assumptions. The ATMS is often used with a problem solver because we need logical variables to represent domain knowledge. APRICOT/0, a hypothesis-based forward-reasoning system, consists of a compiler, the Rete-based inference engine and the ATMS. The compiler translates first-order Horn clauses and normal defaults into a Rete-like network. The Rete-like network consists of a root node,one-input nodes, two-input nodes and terminal nodes. Allowing faster hypothesis-based reasoning, the inference engine gives intermediate justifications to the ATMS at two-input nodes and store the sets of hypotheses where those intermediate data hold. However, the whole of the input knowledge base has to be recompiled and the previous beliefs of the conclusions have to be revised on the whole when the input knowledge base is partially modified by the user. Therefore, both an incremental compiler and an incremental inference system are required. This paper presents an incremental hypothesis-based reasoning system which includes an incremental compiler and an incremental inference system. The incremental compiler is implemented with a modified ATMS which is called RATMS. The RATMS can efficiently offer the inference engine the updated Rete-like network. The incremental inference system can partially revise previous beliefs of conclusions with additional knowledge including variables. The incremental hypothesis-based reasoning system has been applied to a logic circuit design problem. The expermental result shows that beliefs of the solutions are efficiently revised with additional knowledge about datapath design.
- 社団法人人工知能学会の論文
- 1991-07-01
著者
関連論文
- 設計問題向けツール・アーキテクチャ (「第五世代コンピュータ」)
- モデル生成型定理証明器を用いたアブダクションの計算における効率化手法
- モデル生成に基づく並列アブダクション
- モデル生成に基づく並列アブダクション
- ノンホーンマジックセット法と関連性テストとの等価性
- 上昇型定理証明の探索効率を高めるノンホーン・マジックセット
- 段階的前向き仮説推論システム
- ATMSを用いた前向き仮説推論システムにおける効率的な推論方式
- ATMSによるRete-Likeネットワークの逐次構築
- 配送計画を対象とした分散協調型問題解決実験システム