AI: 人工智慧導論課程綱要

Download Report

Transcript AI: 人工智慧導論課程綱要

AI: 人工智慧簡介








人工智慧(Artificial intelligent, AI) 簡介
問題與求解
知識表示方法
推理機制與專家系統
類神經網路
模糊控制
基因演算法
智慧型代理人
人工智慧
 以電腦或機電整合裝置協助或代替人類解決
日常生活或專業領域所面臨的問題。
 其探索領域包括邏輯、機率和連續數學、感
知、推理、學習和行動
 通常以智慧型代理人來實現人工智慧之理念,
可從環境中接收感知資訊,在以擬人化解決
問題之眼算法來執行對應行動。
Intelligent machines, or what machines
can do
 Philosophers have been trying for over 2000 years
to understand and resolve two Big Questions of the
Universe: How does a human mind work, and
Can non-humans have minds? These questions
are still unanswered.
 Intelligence is their ability to understand and learn
things.
Intelligent machines, or what machines
can do
 Intelligence is the ability to think and understand
instead of doing things by instinct or automatically.
(Essential English Dictionary, Collins, London,
1990)
 In order to think, someone or something has to
have a brain, or an organ that enables someone or
something to learn and understand things, to solve
problems and to make decisions. So we can define
intelligence as the ability to learn and understand,
to solve problems and to make decisions.
 The goal of artificial intelligence (AI) as a science
is to make machines do things that would require
intelligence if done by humans. Therefore, the
answer to the question Can Machines Think? was
vitally important to the discipline.
 The answer is not a simple “Yes” or “No”.
Problem Solving Procedures





Defining and Representing the Problem
Descriptions of Problem Solving
Selecting Some Suitable Solving Methods
Finding Several Possible Solutions
Choosing One Feasible Solution to Make a
Decision
Defining and Representing the Problem
 State space: The combination of the initial state and the
set of operators make up the state space of the problem.
 Initiate state: The original state of the problem.
 Operators: used to modify the current state, thereby
creating a new state.
 Path: The sequence of states produced by the valid
application of operators from an old state to a new state
is called the path.
 Goal state: A state fit to the searching objective is
called the goal state.
Common Problem Solving Methods




線性規劃法 (Linear Programming: LP)
整數規劃法 (Integer Programming)
分支界限法 (Branch and Bound Method)
啟發式解法:MST, RMST, SA
 Kruskal, 1956; Prim, 1959; Dijkstra, 1959;
Sollin, 1965; Yu, 1998
 全域搜尋法 (Global Searching):GA
 Brute-Force Search Method
知識表示方法
 Represent and manipulate the domain
knowledge
 Knowledge Definition
 Knowledge Storing
 Knowledge Representation
Knowledge Definition
 “The fact or condition of knowing something
with familiarity gained through experience
or association.” (Webster’s Dictionary,
1988)(Knowing something via seeing,
hearing, touching, feeling, and tasting.)
 “The fact or condition of being aware of
something” .(Ex. Sun is hot, balls are round,
sky is blue,…)
Knowledge Storing
 Natural language for people
 Symbols for computer: a number or
character string that represents an object
or idea (Internal representation of the
knowledge).
 The core concepts: mapping from facts to
an internal computer representation and
also to a form that people can understand.
Knowledge Representation
 Simple facts or complex relationships
 Mathematical formulas or rules for natural
language syntax
 Associations between related concepts
 Inheritance hierarchies between classes of
objects
 Knowledge is not a “one-size-fits-all”
proposition
Knowledge Representation Methods
Effective knowledge representation
methods:
 Easy to use.
 Easily modified and extended (changing
the knowledge manually or through
automatic machine learning techniques).
Knowledge Representation Methods





Procedural method
Declarative method
Relational method
Hierarchical method
Complex network graph
推理機制與專家系統
 Reasoning with rules
 Forward chaining
 Backward chaining
 Rule examples
 Fuzzy rule systems
 Planning
Reasoning with rules
 Rule format: If-then-else rules
 The reasons for using rules




Easily understand
Easily read
Easily add
Easily modify
Reasoning with rules
 Rules are normally manipulated by reasoning
systems:
 Forward chaining: can generate new facts.
 Backward chaining: can deduce whether statements
are true or not.
Rule base system
 An expert system or knowledge-based system
is the common term used to describe a rulebased processing system. It consists three
elements:
 A knowledge base (the set of if-then-else rules and
known facts)
 A working memory or database of derived facts and
data
 An inference engine which contains the reasoning
logic used to process the rules and data.
An example of simple rule
if num_wheel =4 and motor=yes
then vehicle =automobile
Antecedent Clause
Consequent Clause
基因演算法
 基因演算法最早由Holland (1975)所提出,它是
一種隨機全域搜尋之求解技術,可以仿照自然界生
物適者生存之智慧型演化法則,逐步的將解答空間
收斂在很小的範圍內,然後從中選取滿意解。求解
方式乃將可行解以字元串形式表示基因串列,然後
將之置入交配池,作為後續之基因演化之交配或突
變之群體。求解過程屬一重複性的程序,朝向事先
規範的方向(適應函數)重複執行,直到收斂解或
設定條件出現為止。
基因演算求解步驟
 步驟1.隨機產生一母體大小為P之基因串列作為起
始解。
 步驟2.將起始母體之各基因串編碼及計算對應之目
標值。
 步驟3.計算各基因串之適應值(fitness value)。
 步驟4.將最差之基因串捨棄,捨棄之基因串則以最
佳之基因串替代,其中為略小於實數X之整數,R
為事先規劃之基因演算之複製比例。
基因演算求解步驟
 步驟5.進行選擇運算:從目前的母體中選擇群體為P之
基因串,並置入交配池中。
 步驟6.進行交配或突變運算:從交配池中選取一對基因
串來產生子代,選出之基因串不再置回原交配池中。
 步驟7.將產生之子代置入新的母體群,若新的母體未滿,
則回步驟6。
 步驟8.確認是否滿足停止準則,假如滿足則停止搜尋程
序,然後從最近一代之基因串中選出最佳之基因串,並
予以解碼,否則繼續產生下一世代。
類神經網路簡介
 「類神經網路是一種由軟體和硬體所組成之
計算系統,使用許多相連的人工神經元來模
仿生物神經網路。透過人工神經元從外界環
境或其它神經元取得所需資訊,經過簡單的
運算之後,將結果輸出到外界環境或其它神
經元,這些資訊可作進階使用。」
類神經網路簡介(續)
生物神經元模型示意圖
神經樹
神經節
神經核
神經軸
類神經網路簡介(續)
人工神經元模型
輸入訊號
處理單元
淨值
連結加權值
轉換函數
閥值
X1
輸出訊號
X2
Wij
Xn
θj netj
f
Yj
類神經網路簡介(續)
 人工神經元資訊處理模型
Y j  f ( Wij X i   j )
i
 其中
 Yj 表模仿生物神經元模型的輸出訊號
 f 表模仿生物神經元模型之轉換函數(transfer function)
 Wij 表模仿生物神經元模型之神經節強度,又稱為連結加
權值
 Xi 表模仿生物神經元模型之輸入訊號
 θj 表模仿生物神經元模型之閥值
類神經網路簡介(續)
類神經網路組成單元
輸入層:用來輸入外在環境之資訊
隱藏層(有些網路模型無此層) :
提供類神經網路表現處理單元間之交互作用,
以及所處理問題之內在結構能力
輸出層:用來輸出訊息給外在環境
類神經網路簡介(續)
倒傳遞類神經網路模型
輸入層
輸
入
向
量
隱藏層
輸出層
輸
出
向
量
類神經網路之類型
 可綜合為四大類
 監督式學習模型(Supervised learning network)
 非監督式學習模型(Unsupervised learning
network)
 聯想式學習(Associated learning network)
 最適化應用網路(Optimization application
network)
類神經網路之應用
實例一:信用卡持卡者之信用判斷
實例二:從技術面預測證交所隔日收盤指數之
變化率
實例三:從技術面預測證交所隔日收盤指數之
變化
類神經網路之應用
實例四:從穩定液之成分及濃度來預估黏度
實例五:從取得之樣本值來診斷乳癌之良惡性
情形
實例六:依據施工條件及地層基礎選擇適合的
建築基樁型式