机器学习基石 - Regularization

机器学习基石下 (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)

Regularized Hypothesis Set

• step back = constrained optimization of Ein$E_{in}$

• Constraint

  

Weight Decay Regularization

• The Lagrange Multiplier

  

• Augmented Error
  
  

• weight-decay regularization

  

• Legendre Polynomials 勒让德多项式

  其正交性是任意两个式子相乘，在[-1, 1]上的积分为0

Regularization and VC Theory

• 不懂啊

General Regularizers

• 几种方法

• L2 and L1 Regularizer

• The Optimal λ$\lambda$