You succeed in applying Regularization to fully connected layers like here,
|
self.dw = (1/m) * dZ @ self.X.T + (Lambda/m) * self.w # dw = dZ * X.T+(lambda/m) * w |
but you failed in doing so in Convolution Layers, you returned the norm of the weights with zero and didn't use lambda in your backward function.
You succeed in applying Regularization to fully connected layers like here,
Deep-Learning-framework/mshtensorflow/Layer.py
Line 25 in 7904b99
but you failed in doing so in Convolution Layers, you returned the norm of the weights with zero and didn't use lambda in your backward function.
Deep-Learning-framework/mshtensorflow/conv_layer.py
Line 55 in 7904b99