Course 3 Week 1 quiz has many questions on topics that are covered in week 2

Course Q&A Deep Learning Specialization Structuring Machine Learning Projects
, ,
1

Hi,

The Course 3 week 1 quiz had many questions on topics (related to train/dev/test tests) that were covered in week 2 lectures. I was wondering why I struggled with the quiz, went back and viewed the videos again related to week 1, but could not find answers.
When viewing week 2 videos, those questions were answered in the lectures.
Please correct this. This is wasting time.
Also, in week 1 questions I have seen questions related to human-level performance, based on the performance of the ornithologists and regular citizens. But, that question, or the previous questions, did not list the details of the performance of ornithologists or the citizens. I had to guess the answer based on the content in the options.

2

Thanks for your observations.

3

I really don’t remember any “forward references” of the type that you are describing. It is definitely true that the quizzes in both weeks are quite challenging. And you definitely know more and are more comfortable with all the issues once you finish Week 2. So it is plausible that the Week 1 quiz would seem easier after completing the whole course, even without any actual forward references.

Prof Ng makes this general point about the idea of “human performance” and he points out that there are many gradations of human performance on a data classification task. I recall that he gave an example of analyzing medical images looking for tumors or other anomalies. In that case you have lots of levels listed in order of increasing performance (accuracy):

  1. The general public would not be good at this at all.
  2. A medical student who might be actually not be a radiologist, but who had done a radiology rotation during their residency, would be better than the general public.
  3. An individual doctor who is a trained radiologist would be better still.
  4. The greatest skill level would be if you assembled a panel of 8 or 10 of the nation’s leading radiologists, recognized by their peers as experts and have them all examine the images.

So in the case of the quiz questions you refer to, you don’t need to know the exact percentage of increased accuracy at bird identification that an ornithologist would have over a member of the general public or even an amateur bird watcher. The point is just that you have an ordering:

general public < amateur bird watcher < trained ornithologist

That’s all you need to understand to answer the question.