# 2017/09/12~13 Machine Learning

## The AlphaGo Phenomenon

Three things you need to use in Deep learning:
* Computer vision
* Speech recognition
* Text mining

## AlphaGo vs Deep Blue

### AlphaGo
台師大黃世傑開發
* Go game (19*19 grid board)
* **General purpose algorithm**
* Not a set of handcraft rules (?)
* Build a dynamic model(?)
* Modular system combining planning and pattern recognition(?)
* Search space(? )
b^d : b = 250, d = 150

### Deep Blue
台大電機系許峰雄在CMU開發Deep thought(深思)，在進入IBM後持續發展出 Deep Blue (深藍)
* 1997 beat chess world champion
* Search space
b^d : b = 35, d=80

* Common
Tree search
Position evaluation funtion

## The Era of Data science and artificial intelligence

* 1PB = 1000 Terabytes = 1000,000 Gigabytes
* Big Data spans four dimensions
* Volume(大量的資料）: 企業充斥著相當大量的資料（terabytes....even petabytes)，舉例來說：可以將 12 terabyte's tweets 用來作對某項產品的情緒分析
* Velocity (處理速度快，能產生資料同時分析資料)，舉例來說：仔細檢查每天創造的500萬次貿易活動，以確定潛在的欺詐行為
* Variety (多樣化，各行各業都能應用) :Big data is any type of data - structured and unstructured data. New insights are found when analyzing these data types together. For example, 監控100個監控攝像機的實時視頻信息，以達到目標感興趣點.
* Veracity(精準性)：How can you act upon information if you don’t trust it? Establishing trust in big data presents a huge challenge as the variety and number of sources grows.
* From model-driven to data-driven approaches (?)
* Data is the king. 演算法+大量資料.Who owns data shall win the battle.

### What's artificial intelligence?
* A.I. is the study of how to make devices do things at which, at the moment, people are better (Rich & Knight, 1991)
* 5thc B.C. Aristotelian logic invented
* 1642 Pascal adding machine
* 1694 Leibnitz reckoning machine


* Acting Humanly
* The Turing Test: Turing (1950) “Computing machinery and intelligence”: “Can machines think?” -> “Can machines behave intelligently?”. An operational test for intelligent behavior
* Suggested major components of AI: (4+2)
* Knowledge representation 知識代表 to store what it knows or hears 將資料用成可以理解的東西存起來
* Automated reasoning 推理 to use the store information to answer questions and to draw new conclusions
* Natural language processing to enable communications 實現通訊
* Machine learning to adapt to 適應 new circumstances 環境
* Additional ability if for total Turing Test
* Computer vision to perceive objects 感知物體
* Robotics to manipulate 操縱 object and move about
* Problem: Turing test is not reproducible, constructive 建設性, or amenable 適合 to mathematical analysis(?)

* Thinking Humanly: Cognitive Science
* How to validate 驗證? (but are now distinct from AI 已獨立學派，不歸類於AI這一塊) Requires
* Predicting and testing behavior of human subjects (top-down)
* Direct identification from neurological data (bottom-up)
* CAPTCHA, Different Painting Styles...,etc.

*** 2017/09/18

* Thinking Rationally: Laws of Thought (based on logic or math; therefor, everything could be automated)
* Normative 準則的(or prescriptive 規定的) rather than descriptive 描述的 (e.g.,what people will actually do)
* Aristotle: what are correct arguments / though processes?
* (三段論證) Confucius is a man, and All men are mortal => Confucius is mortal
* Not all intelligent behavior 智慧型回 is mediated 調控 by logical deliberation 邏輯思考

* Acting Rationally
* Rational behavior: doing the right thing that which is expected to maximize goal achievement, given the available information

* Types of knowledge: deduction ↔ induction
* Deduction process: less to more rules (math, physics, ...)
* Induction process: more to less rules (biology, social science, ...), just likecompression

* How to ask machine to do what we can do or do more than we can do?
* Modeling (with an oracle)-> automation
* Learning-> performance is improved

### Mordern AI vs Tradittional AI

### Type of Machine learning algorithms

* Supervised learning(icing): give label for each objects, learn to predict and learn from input data Category. For example, classification, regression
* (x,y): x is the feature and y is the label/class
* could ignore same data
* Unsupervised learning(cake): group several object to same group (unlabeled object), learn to inherent structure and learn from input data. For example, clustering, density estimation
* find some of grouping(?) For example, chairs.
* Semi-supervised learning, a mixed case of the above two! Usually we have much more unlabeled data than labeled data!
* Reinforcement 加強 learning(cherry, decoration), where we only get feedback in the form of how well we are doing (not what we should be doing): human play a game for a while then computer could learn from that. For example,planning

### Two examples of machine learning

## Big data

* 4V
* Data is the king: who owns data shall win the battle

## Thinking