CHANG teacher machine learning course notes _ML Lecture 0-1: Introduction of Machine Learning

introduction:

Recently started to learn "machine learning", the teacher had heard CHANG island country's name, has no time to watch his courses. After listening to a lesson today, I feel great, easy to understand, but can seize the key, but also the middle with some very interesting examples to deepen the impression of the students.
Video Links (bilibili): CHANG Machine Learning (2017)
In addition students have been determined to do shorthand and updates on github: CHANG machine learning notes (LeeML-Notes)
So, the next record my notes just some of my own summary and lectures at the time of confusion, if there is a friend can help me please exhibitions.

1, the relationship between artificial intelligence, machine learning, deep learning of

Artificial intelligence is the goal, machine learning is a means to an end, and the depth of learning is one way to achieve machine learning.
Human wants to create "smart", there are two ways:

• hand crafted rules (rules artificial prepared): requires a lot of if then, is an abandoned route
• So that the machine can own learning: the curriculum is to explore the direction of
  
  Yann LeCun once issued a cartoon on twitter, the irony is that many of the so-called cool appearance AI, open source is actually see a lot of hand crafted rules procedures only. This in turn reminiscent had previously been said, "with PPT to write more often than not artificial intelligence, is written in python." So it seems, is not written in python Insurance ah.
  ### 2, machine learning is to do what?
  
  ≈ seek a machine learning function (function)
  This function can be determined in our input data, after it later, you can output the results we want.
  How ### 3, machine learning?
  
  

How to find the function mentioned above? To image recognition, for example, you want to find the right function, you need to prepare a set of functions, this series set of functions is called model (Model). Model where we need is some function, such as the image above f1, because when you enter cat picture, it can give correct results. And f2 is obviously not what we want function. The purpose of machine learning is to find all the animals recognize the optimal function f *. When in the training data, not only shows the data (picture) input, but also gives the results of (name of animal) we want the output of the machine, the machine learning is called supervised learning. The output is given in supervised learning is also known labels.

4, step machine learning

Machine learning process is often divided into three big (like the elephant in the refrigerator):

• Defines a set of functions, or which model selection
• The machine can measure a function of the quality of
• So that the machine can pick out the best function

5, machine learning learning route

Machine Learning divided (by scene, more and more difficult):

• Supervised learning (by task):
  1. Regression (regression): need is a digital output (scalar), such as data tomorrow forecast of PM2.5
  2. Classification (classificasion):
     • Binary classification problem: the need for output is "yes" or "no", such as discrimination spam
     • Multivariate classification problems: such as automatic classification of news content, are finance, sports, politics ......
  3. Structured learning (structure learning): Like the dark continent, there are many things we do not understand, more well-known method - to generate confrontation Network GAN
• Semi-supervised learning: a small amount of data labels, a large number of unlabeled data
• Transfer learning: in addition to a number of unlabeled data, there are a lot of work to be done has nothing to do with the data
• Unsupervised Learning: learning unlabeled data
• Reinforcement Learning: the need to learn by evaluating, learning from critics, more in line with human learning mode

Classification model which can be used are:

• Linear model: simple
• Non-linear model: where to spend energy major, non-linear model is the most familiar deep learning model (divided into the following methods):
  1. Depth study
     • Image Recognition: Andrew Ng then let the computer use in google depth study identified the cat
     • Go: AlphaGo, similar to a 19 * 19 (Go board) classification problem
  2. K-NN
  3. Decision Tree
  4. SVM (support vector machine)