机器学习基石 - Noise and Error

机器学习基石上 (Machine Learning Foundations)—Mathematical Foundations
Hsuan-Tien Lin, 林轩田，副教授 (Associate Professor)，资讯工程学系 (Computer Science and Information Engineering)

Noise and Probabilistic Target

确定的改为概率，保证同一分布下

练习题

Error Measure

• out of sample
• pointwise
• classification
• 0/1 error often for classification; squared error often for regression
• extended VC theory/philosophy works for most $\mathcal H$ and err

Algorithmic Error Measure

• err is application/user-dependent
• 学习流程图
  

Weighted Classification

• 不同的情形有不同的权重
• Weighted Pocket Algorithm