Artificial Intelligent http mail im tku edu twychou

人 智慧 Artificial Intelligent 淡江大學 資訊管理系所 侯永昌 http: //mail. im. tku. edu. tw/~ychou ftp: //mail. im. tku. edu. tw/Prof_Hou 淡江大學資訊管理系所 侯永昌

知識庫表示法：語意網路 • Nodes︰代表物件，用以表示功能、動 作、狀態、… • Links︰連接相關的nodes，用以表示關 係、轉換、… • Relationship is basic structure for organizing knowledge. • With relationship, other knowledge can be inferred. 淡江大學資訊管理系所 侯永昌 9

語意網路的優缺點 • 優點︰ – Flexibility︰新增node及link都很容易 – Inheritance︰藉由is-a relationship – 減少redundency︰藉由inheritance • 缺點︰ – Flexibility︰沒有什麼作業標準 – Difficult to handle exceptions – Combinational explosion of searching nodes 淡江大學資訊管理系所 侯永昌 12

知識庫表示法：物件-屬性-值 Object apple grapes Attribute color type quantity Value red Washington 100 red seedless 500 • 非常方便將知識建立在表格中 淡江大學資訊管理系所 侯永昌 13

知識庫表示法：物件-屬性-值 • 物件-屬性-值 vs. 語意網路︰ – nodes︰objects, attributes, values – has-a link：object-attribute – is-a link：attribute-value • O-A-V易於表現不確定的資訊 (再加上 一個certainty factor欄位) • Certainty Factor: represent the confidence of a piece of evidence 淡江大學資訊管理系所 侯永昌 14

知識庫表示法：法則 • 通常以if-then的形式呈現 • 例：if tattoo is a dragon, and color of the dragon’s scales is yellow, then origin of tattoo is China. • 有關法則的定義可以是固定的、可改變 的、有條件的、甚至是機率性的 • 例：if tattoo is a dragon, and color of the dragon’s scales is yellow, then origin of tattoo is China. (0. 7) 淡江大學資訊管理系所 侯永昌 16

知識庫表示法：法則 • Rule 1: if tattoo is a fish, and color of the fish’s scales is pink, then origin of tattoo is Japan. Rule 2: if tattoo is a snake, and color of the snake’s scales is blue, then origin of tattoo is Hong Kong. Rule 3: if tattoo is a dragon, and color of the dragon’s scales is yellow, then origin of tattoo is China. • 每一個類似的法則都要列舉，太麻煩了 淡江大學資訊管理系所 侯永昌 17

變數法則(variable rule) • Rule: if tattoo is an X, and color of the X’s scales is Y, and origin (X, Y) is Z then origin of tattoo is Z. • Facts, knowledge: origin (fish, pink) is Japan, origin (snake, blue) is Hong Kong, origin (dragon, yellow) is China, … 淡江大學資訊管理系所 侯永昌 18

• 優點： 法則的優缺點 – Modularity – Uniformity – Naturalness – flexible hierachy • 缺點： – the operation is relatively inefficient – It is hard to follow the control of problem solution – The hierachical relationship are difficult to visualize -- flat 淡江大學資訊管理系所 侯永昌 19

Frames: A Simple Example 淡江大學資訊管理系所 侯永昌 21

框架的缺點 • Allow unrestrained alteration or cancellation of slots as exceptions • Since any slot can be changed, the inheritance properties can be altered or cancelled anywhere in the hierachy. no universal statement 淡江大學資訊管理系所 侯永昌 25

What is Logic? • Reasoning about the validity of arguments. • An argument is valid if its conclusions follow logically from its premises – even if the argument doesn’t actually reflect the real world: – All lemons are blue – Mary is a lemon – Therefore, Mary is blue. 淡江大學資訊管理系所 侯永昌 29

Logical Operators • Not • And • Or • Implies • Iff (if… then…) (if and only if) 愈往上，優先順序愈大 淡江大學資訊管理系所 侯永昌 30

Translating between English and Logic • Facts and rules need to be translated into logical notation. • For example: – It is Raining and it is Thursday: –R T – R means “It is Raining”, T means “it is Thursday”. 淡江大學資訊管理系所 侯永昌 31

Truth Tables • Tables that show truth values for all possible inputs to a logical operator. • A truth table shows the semantics of a logical operator. A A A B B B false false true true true 淡江大學資訊管理系所 侯永昌 32

Translating between English and Logic • More complex sentences need predicates. E. g. : – It is raining in New York: – R(N) – Could also be written N(R), or even just R. • It is important to select the correct level of detail for the concepts you want to reason about. 淡江大學資訊管理系所 侯永昌 33

知識庫表示法：邏輯 • Predicate P(x 1, …, xn) is the statement about objects that is used to represent relationship • The assertion of predicate is either T or F • 例如： weather(Tue, rain) likes(Mary, George) ( x) (human(x) mortal(x)) 淡江大學資訊管理系所 侯永昌 35

知識庫表示法：邏輯 • 敘述計算學(Predicate Calculus)：用來處 理predicates之間推論的邏輯。通常利用 替代(substitute)、匹配(matching)和一致 化(unification)來處理predicate中有關變 數的推論 • 一致化：the process of finding substitutions for variables to make arguments match is called unification 淡江大學資訊管理系所 侯永昌 36

Quantifiers and • - For all: – x P(x) is read “For all x’es, P (x) is true”. • - There Exists: – x P(x) is read “there exists an x such that P(x) is true”. • Relationship between the quantifiers: – x P(x) ¬( x) ¬P(x) (O) – x L(x, C) 淡江大學資訊管理系所 侯永昌 – x L(x, C) (O) (X) 38

Equivalence • Two expressions are equivalent if they always have the same logical value under any interpretation: – For example: A B B A • Equivalences can be proven by examining truth tables. 淡江大學資訊管理系所 侯永昌 39

Some Useful Equivalences • • • A A (B C) (A B) A (A B) A A true A B (A B) A B A C C (A (A A C) C) (A B) A false A (A B) A B 淡江大學資訊管理系所 侯永昌 40

Properties of Logical Systems • Soundness: Is every theorem valid? • Completeness: Is every tautology a theorem? • Decidability: Does an algorithm exist that will determine if a wff is valid? • Monotonicity: Can a valid logical proof be made invalid by adding additional premises or assumptions? 淡江大學資訊管理系所 侯永昌 41

Modal logic • Modal logic is a higher order logic. • Allows us to reason about certainties, and possible worlds. • A statement that may be true or false, depending on the situation, is called contingent (偶然的). 淡江大學資訊管理系所 侯永昌 43

Abduction and Inductive Reasoning • Abduction: B A→B A • Not logically valid, BUT can still be useful. • In fact, it models the way humans reason all the time: – Every non-flying bird I’ve seen before has been a penguin; hence that non-flying bird must be a penguin. • Not valid reasoning, but likely to work in many situations. 淡江大學資訊管理系所 侯永昌 44