Question 9 asks us to check all answers about CNN that are correct. One of the answers says CNN allows transfer learning and I picked that. But the grader says no, I shouldn’t have chosen that as part of the right answers because “transfer learning is not bound to convnet”. I’m confused. This question doesn’t ask what properties are only specific to CNN, rather it asks which of the given facts are correct about CNN. For that wording, the answer that includes “CNN allows transfer learning” should be included as part of the right choices. Because of this stupid error from the grader, I had to redo the quiz again! Seriously. What a joke.
Hello Zhexingli,
Rightly said.
@Mubsi, I think Zhexingli has pointed it out correctly. The wordings of the question must be paraphrased again.
DLS_Course 4_Basics of Convnets_Graded Quiz_Q9
Yes, the wording there is perhaps a little confusing. What they meant is that this question is about the specific meaning of “parameter sharing” in ConvNets as Prof Ng defines it in the lectures. Yes, “transfer learning” can apply with ConvNets, but we don’t call that “parameter sharing”. Of course you can argue this is nitpicking, but that’s not what they call it in the transfer learning case.
There were a couple of earlier questions on the quiz that showed the difference in the number of parameters in an FC net handling 300 x 300 color images with 100 neurons versus the number of parameters in a ConvNet handling the same size images with 100 5 x 5 filters and the comparison is pretty stark. Between 3 and 4 orders of magnitude basically. That’s what parameter sharing is referring to.
How do we open an issue for it here?
Hello @zhexingli! You don’t need to open an issue for it. We (mentors) do this, in this case, @Rashmi will do it.
Best,
Saif.
Hello Zhexingli,
I have already created a new issue for this query on Github.
@Mubsi and the staff will take care of it and will be resolved at the earliest.
1 Like