深度学习 初始化、正则化

1.初始化 initialization

A well chosen initialization can:

• Speed up the convergence of gradient descent
• Increase the odds of gradient descent converging to a lower training (and generalization) error

要点笔记：
以三层神经网络为例

• Zeros initialization

• Random initialization
  This initializes the weights to large random values.

      parameters['W' + str(l)] = np.random.randn(layers_dims[l], layers_dims[l-1]) * 10
      parameters['b' + str(l)] = np.zeros((layers_dims[l], 1)) 
  

随机初始化的数值过大时，效果不好
那么怎么确认改初始化多大的值呢？
看下面这个： He initialization

• He initialization
  （This initializes the weights to random values scaled according to a paper by He et al., 2015.）
  与 "Xavier initialization"类似，只是 Xavier initialization用了np.sqrt(1/layers_dims[l-1]
  而He initialization 用的是np.sqrt(2/layers_dims[l-1]

parameters['W' + str(l)] = np.random.randn(layers_dims[l], layers_dims[l-1]) * np.sqrt(2/layers_dims[l-1])
parameters['b' + str(l)] = np.zeros((layers_dims[l], 1)) 

W需要随机初始化，b不怎么需要.

---

2.正则化 Regularization

避免过拟合的常规方法是 L2 regularization.它修改了代价函数,
from:
J = − 1 m ∑ i = 1 m ( y ( i ) log ⁡ ( a [ L ] ( i ) ) + ( 1 − y ( i ) ) log ⁡ ( 1 − a [ L ] ( i ) ) ) (1) J = -\frac{1}{m} \sum\limits_{i = 1}^{m} \large{(}\small y^{(i)}\log\left(a^{[L](i)}\right) + (1-y^{(i)})\log\left(1- a^{[L](i)}\right) \large{)} \tag{1}