SEHH3326 Introduction to Artificial Intelligence and Machine Learning
喱一科真係好有趣 :D , 不過睇真 D 原來係 Discrete Math >:(
Table of Content
- Ch1 Introduction to AI and ML
- [Ch2 Searching]
- [Ch3 Adversarial Search]
- [Ch4 Logical Agents and Reasoning]
Ch1 Introduction to AI and ML
咩係 AI 同 ML?
i. Artificial Intelligence (AI)
對於 AI 一詞，唔同學者都有唔同 ge expectation
Psychologist:[Human Peformance/Thought/Process] vs Engineer:[Rationality/Behaviour/Result]
psychologist (心理學家)認為 AI 應該要 behaves like a human (Human performance)
Computer Engineer 認為 AI 應該要可以 efficiently, quickly, cheaply 咁 achieve the goal (do the right things) (rationality 理性)
但係大多數 ge AI researcher 都係注重係 result 多個 process
<img src=”https://i.imgflip.com/67v1zt.jpg” width=40% height=50%>
- Intelligence
- 指識得 Learning (學習), Reasoning (推理), Understanding (理解) environment ge 野
- Artificial Intelligence (AI)
- 指一個可以 show 到 Intelligence 特性 ge 一個由人所 build ge entities
- Computer programming vs AI
- 係 computer ge 世界入面，你要知道點樣去 solve 嗰 problem，再 step by step (programming) 咁教電腦
- 係 AI ge 世界，你唔洗知點樣去 solve 嗰 problem 而係話比 AI 知你想要 D 咩
- AI vs ML
- AI 係用 reasoning, thinking 同 logics 去 solve problems
ML 係用大量 example 去 solve, 同埋識進化 (improve)
- Turning Test
- 一個攞喱 check AI 夠唔夠 human ge test，但係其實都幾難 pass 到
ii. History of AI
諗到 AI 喱一樣 ge 人真係好 big brain 早期 ge AI 因為 computing power ge 不足，一直都受到限制
- Expert systems
早期類似 AI 概念 ge 一個 system
- 由一大堆 if then statment 所組成
- 唔識進化
- 理解唔到新野
- 好難 maintain
- Late 1980’s
- Probabilistic reasoning (probability + statistic)
- Machine learning
- Early 2000’s
- Big Data
- Early 2010’s
- Deep Learning
iii. Application of AI
iv. Limitation of AI
盡管 AI 睇好似似萬能咁，但係其實 AI 都有唔擅長 ge 野 語言博大精深
- E.g. Language translation, story telling
- The rule of language often contradict itself
- A single word can have various meaning
- New word being created constantly
v. Intelligent Agents
一個 agent 係由 Architecture 同 agent program 所組成 為左可以整到一個 Rational agent，我地需要一 D performance measur，一個 environment，再加上兩類 architecures
- Performance measure
- Environment
- Actuators (執行)
- Sensors
而當中 Environments 入面又有分唔同 types
為左方便去認，以下會以 player 代指 agent
- Fully vs Partially observable
- 就環境而言，我地係咪知道哂所有資訊?
- Single-agent vs Multiagent
- 得一個 player? 定係仲有其他player?
- 雙方可以係 competitive or cooperative ge 關係
- Deterministic vs Nondeterministic
- 當 player 做出一個動作嗰時，會唔會知道嗰結果?
- Episodic vs Sequential
- player 所做 ge 每一個 move 係獨立 ge? 定係因影響之後 ge move?
- Discrete vs Continuous
- 周圍 ge 環境，時間，規則同動作係一格一格咁跳，定係連續咁喐動?
- Jason 講到一個例子：如果嗰 environment 係用電子鐘 ge 話，咁就係discrete，但係如果係用秒針一直喐 ge 嗰隻鐘(唔係一秒一秒跳嗰 D)就係 continuous
- Static vs Dynamic vs Semidynamic
- Static 指，環境會等到 player 做出一個動作之後先會改變
- Dynamic 指，環境係會係 player 諗緊嗰時都可以做出改變 (例如對手做左一個動作都計改變左嗰環境)
- Semidynamic 指，環境雖然唔係會 player 做出動作之前改變，但 player ge performance score 卻可以
- Known vs Unknown
- 哩一項並唔關嗰 environment 事，而係關 player 本身
- Known enviroment 係指 player 知道會一個動作 ge outcome (or probability outcomes) 係咩
- Unknow environment 就係 player 知佢可以做咩，但唔知嗰 enviroment 會點變
除左關於 environment 之外，仲有關於 agent programs
agent program 可以話係 Brain of agent 最常見 ge 有以下四種
- Simple reflex agents
一種淨係會 focus 係當下環境而作出決定 ge agents
- 靠 if then rules 去做決定
<img src=”https://media.discordapp.net/attachments/684958583367925771/951021882923175956/unknown.png” width=50% height=50%>
- 靠 if then rules 去做決定
- Model-based reflex agents
一種會 base on 當下 ge environment 同過住經驗去做決定 ge agents
- agent 會保持住一種叫 internal state ge state
- internal state 係由 Transition model 同 Sensor model 去作 update
- Transition model 會話比 agent 知嗰環境係點變 (may or may not depend of agent’s actions)
- Sensor model 會話比 agent 知嗰環境會點樣影響嗰知覺
- Goal-based agents
- <img src=”https://media.discordapp.net/attachments/684958583367925771/951025320620863498/unknown.png” width=50% height=50%>
- Utility-based agents
- <img src=”https://media.discordapp.net/attachments/684958583367925771/951025429525979156/unknown.png” width=50% height=50%>