Cousera机器学习基石第三周笔记 Machine Learning Foundation Week 3 Note in Cousera

Types of Learning

Learning with Different Output Space

Credit Approval Problem Revisited

More Binary Classification Problems

Multiclass Classification: Coin Recognition Problem

Regression: Patient Recovery Prediction Problem

Structured Learning: Sequence Tagging Problem

Learning With Different Data Label

Supervised: Coin Recognition Revisited

## Unsupervised: Coin Recognition without \(y_n\)

unsupervised multiclass classification \(\Leftrightarrow\)‘Clustering’:a challenging but useful problem

Unsupervised: Learning without \(y_n\)

  • Clustering
  • Density estimation
  • Outlier Detection

Semi-supervised:Coin Recognition with Some \(y_n\)

avoid expensive labeling

Reinforcement Learning

Learning with Different Protocol

Batch Learning

a very common protocol

Online: Spam Filter that ‘improves’

Active Learning: Learning by ‘Asking’

Learning with Different Input Space

Concrete Features

human intelligence

Raw Features

need human or machines to convert to concrete ones(feature engineering)

Abstract Features

again need feature conversion/extraction/construction

Summary

  • Learning with Different Output Space:classification,regression,structured
  • Learning with Different Data Label\(y_n\):supervised,un/semi-supervised,reinforecement
  • Learning with Different Protocol:batch,online,active
  • Learning with Different Input Space:concrete,raw,abstract