Week2- programming assignment- mobile net- activation funcation

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Hi, For the alpaca model , the activation function used is ‘linear’. Is there any reason why we have not used ‘sigmoid’?

2

Because

model2.compile(optimizer=tf.keras.optimizers.Adam(lr=base_learning_rate),
               loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
               metrics=['accuracy'])

we specify from_logits=True here, which effectively means you end up using the sigmoid activation function in the last layer, but it is optimized.

3

Yes, I understood it now. Thanks!