[Coursera] Stanford Machine Learning

01. Introduction

Machine Learning

• Grew out of work in AI
• New capability form computers

Examples

• Database mining
  • Large datasets from growth of automation/web.
  • E.g., Web click data, medical records, biology, engineering
• Applications can't program by hand.
  • E.g., Autonomous helicopter, handwriting recognition, most of Natural Language Processing(NLP), Computer Vision.
• Self-cusomizing programs
  • E.g., Amazon, Neflix product recommendataions
• Understanding human learning (brain, real AI)

02. What is machine Learning

Machine Learning definition

• Arthur Samuel (1959)

Machine Learning: Field of study that gives computers the ability to learn without being explicitly programmed.

• Tom Mitchell (1998)

Well-posed Learning Problem: A computer program is said to learn from experience E with respect to some
task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.

Machine learning algorithms:

• Supervised learning
• Unsupervised learning
• Others: Reinforcement learning, recommender systems.

03. Supervised Learning

• "right answers" given

Regression (회귀)

• Predict continuous valued output(price)
• 연속적인 결과값을 예측

Classification (분류)

• Discrete valued output (0 or 1)
• 0 또는 1, 양성인지 악성인지 등과 같이 불연속적인 결과값을 예측

04. Unsupervised Learning

• 클러스터링 알고리즘 (ex 구글 뉴스)

• 비지도 학습의 예

• Orhanize computing clusters
• Social network analysis
• Market segmentation
• Astronomical data analysis

Cocktail party problem

• 두 개의 소리가 녹음된 마이크에서 하나의 소리를 분리해내는데는 단 한 줄의 코드면 충분하다.
  

  

Octave

• Octave나 Matlab 같은 도구를 사용하면 많은 학습 알고리즘을 몇 줄의 코드로 구현할 수 있다.

05. Linear regression with one variable Model representation

• 집세에 대한 예측 (Housing Prices)

• Supervised Learning
  • Given the "right answer" for each example in the data.
• Regression Problem
  • Predict real-valued output
• Classification

Training set (학습 데이터)

• Notation
  • m = Number of training examples
  • x's = "input" variable / features
  • y's = "output? variable / "target" varaible
  • (x, y) - one training example
  • (x(i), y(i)) - i th training example

선형회귀

• 단일변량 선형회귀
• 단일변량 : 하나의 값

Cost function (비용함수)

• We can measure the accuracy of our hypothesis function by using a cost function.

• 비용함수를 사용하면 주어진 데이터에 가장 가까운 일차함수 그래프를 알아낼 수 있다.

• 이 함수를 "제곱 오차 함수" 또는 "평균 제곱 오차"라고 한다.

Cost Function - Intuition I

Cost Function - Intuition II

Gradient descent

• 기울기 하강은 기계학습의 모든 곳에서 실제로 사용되고 있음
• 비용함수 j의 최소값을 구하는 알고리즘