Week01. Course Introduction

Course_Lin_MLF

1. Course Design

• theory oriented
• techniques oriented
• our apporoach
  foundation oriented
  
  • story-like
    When can Machines learn?
    Why ...?
    How...?
    How... work better?
  • mandarin teaching
    8 week + 7 week

2. What is Machine Leaning ?

• From Learning to ML
  man learning:$\quad observations \to \mathbf{learning} \to skills$
  machine learning: $\quad data \to \mathbf{ML} \to skills$

• A More Concrete Definition
  skill: improve some performance measure
  machine learning: $\quad data \to \mathbf{ML} \to improved \ performance \ measure$

• The Machine Learning Route
  ML: an alternative route to build complicated systems

  • 应用场景
    当无人类在场，无法编程
    人类不能轻易定义或得出解
    当需要快速的决定时(远超出人类的决定速度)
    when needing to be user-orieted in a massive scale
  • not give a computer a fish, but teach it how to fish

• Key Essence of Machine Learning
  1) exist some 'underlying pattern ' to be learned
  存在pattern可以学
  2) but no programmable (easy) definition
  3) there is data about the pattern

3. Application of Machine learning

• Daily Needs:Food,Clothing,Housing,Transportation
• Education
• Entertainment: Recommender System

4. Components of ML

• Formalize the Learning Problem

  • Basic Notations
    $$\begin{align*} input: x \in \mathcal{X} \\ output: y \in \mathcal{Y} \\ unkonw \ target \ function: f: \ \mathcal{X} \to \mathcal{Y} \\ hypothesis: g: \ \mathcal{X} \to \mathcal{Y} \\ \end{align*}$$
  $$\begin{align*} hypothesis \ set: \ \mathcal{H} \\ learning \ algorithm: \ \mathcal{A} \end{align*}$$
  $learning \ model = \mathcal{A} \ and \ \mathcal{H}$
  $\mathcal{A} \ takes \ \mathcal{D} \ and \ \mathcal{H} \ to \ get \ g$
  • Practical Definition of ML
    machine learning: use data($\mathcal{D}$) to compute hypothesis $g$, that approximate target $f$.

5. Machine Learning and Other Fields

• Machine Learning and Data Mining(　资料探勘(台)　)
  Machine Learning
  Data Mining: use (huge) data to find property that is interesting

  • if 'interesting property' same as 'hypthesis that approximate target'
    ML = DM
  • ... related to ...
    DM can help ML, and vice versa
    difficult to distinguish ML and DM in reality

• ML and AI
  Artificial Intelligence: compute something that shows intelligent behavior

  • $g \approx f$ is somthing that show intelligent behavior
    ML can realize AI, among other routes
    e.g. chess playing

• ML and Statistic
  Statistics: use data to make inference about an unkown process

  • $g$ is an inference outcome; $f$ is something unkown
    statistics can be used to achieve ML
  • traditional statistics...
    statistics: many useful tools for ML