1. Introduction to Machine Learning

Era of Artificial Intelligence

• Personalized item Recommendation

• Self-Driving Car

• What is a machine?

  Related to Model

• From what does a machine learn?

  Related to Training data

• How does a machine learn?

  Related to Learning Algorithm

What is Machine?

Hardware or software to perform an intended function

$$f(x) = y$$

Real examples

How to make?

Traditional paradigm

• Writing a program code

• Problem

  We do not have a clear description about the function that we want.

  

Current paradigm

• Letting the machine learn the function(pattern) from observed data. Machine finds pattern from dataset.

  

Definition of dataset

A set of samples of mapping between input and output

Requirements

1. dataset

2. model about function

• Example : Linear model

  

  $\tau : threashold$

  Learn 3 parameters. Fine the optimal value of three parameters. ($w_{age}, w_{weight}, \tau$)

How does model can learn?

Learning process aims to seek the best function that minimizes the error

Three types of Machine Learning

Supervised learning

Each example in a dataset is labeled in advance (Labeling is expensive)

Unsupervised learning

A dataset has no labels. This type of learning highly popular in modern days because labeling is so expensive. Actually, combining supervised learning and unsupervised learning is the mainstream of the industry.

• Self-supervised learning

  Create a label with data itself.

Reinforcement learning

Its label is a real-valued reward signal(possibly delayed).

• Problem

  Difficult to define the label.