Regularization derivative for L2 norm

Course Q&A Deep Learning Specialization Improving Deep Neural Networks: Hyperparameter tun
,
1

image
image1096×519 86.2 KB

when it talks about the formula for dw as dW[L]=(from backprop) + lambda/m * W[L]. I was wondering why it was not + lambda/2m instead. This is because it was shown in the formula but not in the derivative.

2

The derivative causes the ‘2’ term to be canceled. It’s an application of the power rule for taking derivatives.