一般化分類者系統-以乏晰分類者為實作

Abstract

所謂知識擷取是指由人類本身, 或由外在環境學習到我們人類能夠理 解, 並加以檢驗的知識. 這在符號人工智慧中, 要對一個人類 能理解的知識表達法 (通常是複雜的)加上學習方式, 或是將數值人 工智慧的機器式知識表達法轉化 成可理解的型式, 這兩者都是非常 困難的. 在人工智慧中研究 中, 整合不同的系統是個漸受重視的問題. 在許多不同的技 術中取得 協調的混合式模型能夠顯現出個別模型都不及的能力, 本論文結合了 人工智慧的兩大分支, 符號人工智慧 ,和數值式人工智慧中的四種不同的 技術: 類神經網路, 遺傳演算法, 乏晰邏輯, 和分類者系統, 成為一 個新的模型,用來 解決在以規則為基礎的系統中, 知識擷取的難題. 並在乏晰邏輯中實作. 本文並改良另一種類神經網路和乏 晰邏輯混合式模型 並以鳶尾花的分類來比較 兩種模型所擷取的知識. Knowledge acquisition means to learn knowledge from human or environment which can be reexamined and verified by human. In symbolic AI, to design a learning mechanism on human-like knowledge representation, which is usually complex, is difficult. On the other hand, to convert internal machine knowledge representation used by numerical AI systems into understandable form is also a challenging task. In the AI field, hybrid models of different systems have become a more and more important topic. A hybrid model which combines and coordinates many different techniques can show individual power behind each component techniques and compensate their drawbacks. This thesis combines four different techniques in two major branches of AI, symbolic AI and numerical AI, they are: neural networks, genetic algorithms, fuzzy logic, and Holland's classifier system. We use this hybrid model to solve the knowledge acquisition problem of a rule-based system. We implement it in a fuzzy logic system. This thesis also improves an existing neuro-fuzzy hybrid model. We compare the knowledge learned by the two models in an iris clustering problem.