[ML] 기계학습 기초(2021) (SNU Machine Learning Fundamentals & Applications)

서울대 딥러닝 유튜브 채널의 강의(윤성로 교수님)를 보고 정리한 내용들.

Topic 1.

• Introduction: Learning from Data
  • What is a tree? - Learning from data
  • Example: predicting how a viewer will rate a movie
  • Example: human-level control (game play)
  • Example: autonomous vehicles
  • The essence of learning from data - Exercise
• Introduction: Problem Setup
  • Running example: credit approval
  • Components of learning: formalization
  • Example - Components of learning: learning algorithm
  • Basic setup
• Introduction: A Simple Learning Model
  • Learning model: solution components
  • Hypothesis set
• Introduction: Perceptron
  • Making a decision
  • The `perceptron'
  • Two-dimensional case
• Introduction: Perceptron
  • Perceptron learning algorithm (PLA)
  • How PLA works
  • Iterations of PLA
• Types of Learning
  • Learning paradigms
  • Supervised learning
    • Classification
    • Regression
    • training(비쌈) -> testing, inference(싸다)
    • (input, correct output)
  • Reinforcement learning
    • Training examples do not say what to do
    • Target output 없음.
    • Some possible output은 있음, 어떤 출력이 얼마나 좋은지 평가하는 grade or reward 있음.
    • (input, some output, grade for this output)
    • 축구에서 패스하는 행위에 대한 평가 같은 느낌.
    • Learning game
  • Unsupervised learning
    • (input,?)
    • 접근 방법들.
      • Clustering(k-means, mixture models, hierarchical)
      • Density estimation
      • Feature extraction ( PCA, ICA, SVD)
    • Semi-supervised learning, self-supervised learning(BERT)
    • Finding PATTERNS and structure
    • 지도 학습의 전 단계처럼 사용.
    • Higher-level representation of the data
      • Automated feature extraction
  • Example: coin clustering problem
  • Unsupervised learning can be viewed as ...
  • Exercise
  • Summary
    • Learning은 데이터에 패턴이 있을 때, 사용.
    • Leaning model: hypothesis set and learning algo
      • Perceptron and PLA
    • 머신 러닝 타입
      • 지도 학습
      • 비지도 학습
      • 강화 학습
    • Supervised learning: main theme
      • Unknown target function y = f(x)
      • Known training data set (x, y) …
      • Learning algo picks g = f from hypothesis set H

참조 링크