Lec9-Normalization, Dropout, + Implementation
Contents
Normalization and Regularization
Normalization and Initialization
注意看weight variance的曲线,几乎不变
norm的思想来源
- layer normalization
- batch normalization
这么看来batch_norm确实很奇怪, odd! 😢
Regularization
L2 Regularization
针对的是过拟合?但是只要是减少function class的操作都是regularization的一种
然后发现weight decay和regularization有联系!
dropout