Semi-supervised
Intro
supervised learning：
Semi-supervised learning：
Why？ 利用好未标注信息
Why helps？ 假设影响效果
Generative Model
类似EM，逐步逼近
Low-density Separation
self training
hard label有绝对的类别，soft label为各类别概率
Entropy-based
Semi-supervised SVM
Smoothness Assumption
Definition
Application
Approach
Cluster
Graph-based
Better Representation
找到本质信息
