Questions I don't think are right on a quiz

Course Q&A Deep Learning Specialization Convolutional Neural Networks
,
1

Week four quiz

I would like to point out that the last couple of layers of a Siamese network seem to be trainable. Citation-- paper given in lecture Specific quote:

“The network has roughly the same number of parameters as the original one, since much of it is shared between the two replicas, but requires twice the computation. Notice that in order to prevent overfitting on the face verification task, we enable training for only the two topmost layers. The Siamese network’s induced distance is: d(f1 f2) = i if1[i] f2[i] , where i are trainable parameters. The parameters of the Siamese network are trained by standard cross entropy loss and backpropagation of the error.”

Style cost function citation from lecture

The style cost function is defined as the Frobenius norm of the correlations between layers in the generated and style images. That can be rephrased as, basically, the inner product of the difference between the style and generated image correlations between layers, which are the inner products of the pixels for each layer times the outer product of the layers. I think there are two sentences on this form that represent that to one degree or another.

2

A few points to consider:

The quizzes are like the assignments in that:

a) We can’t publicly share the answers or even the questions in the case of the quizzes.
b) We can’t see your work, so your link above just takes me to the quiz and I can take it myself.

Also note that there are multiple questions available in each category, so you don’t necessarily see the same questions each time you take the quiz. I took the C4 W4 quiz a couple of times and did not see any questions that asserted anything about the trainability of Siamese networks, but maybe I just got unlucky.

Some of the questions are a bit on the subtle side and can be subject to misinterpretation, but it’s always possible that they’ve added some new questions that are in fact incorrect.

If you want to go into more detail about any particular questions, we’ll need to do that on a DM thread.