-
is it possible to get correct answers for questions answered incorrectly?
-
Q3 - Recall the diagram of iterating over different ML ideas. Which of the stages shown in the diagram was improved with the use of a better GPU/CPU? Of the 4 answers I believe only one is correct (running experiments - executing code faster). Answers 2 and 3 are also true in general but they are not unique improvements due to use of better GPU/CPU. And #4 that w/o better h/w no way to train faster is not true. If you agree please correct the grading,
-
Q 8 - Why can an RNN (Recurrent Neural Network) be used to create English captions to French movies? Choose all that apply. Here too I would argue that though two answers are true about RNN’s in general, the one feature uniquely relevant to creating captions is that RNN can map a sequence to a sequence. There are other supervised learning algorithms but they cannot do this. If we focus on the essence of the question and look for truly discriminating answer then only one would be true. if you agree then please update the grading.
There are no “official” correct answers that are published. Just as there are no “official solutions” to the programming exercises. If there were, it would just be a source of rampant cheating.
For Q3, I don’t have the quiz in front of me, so I have to take your word for choices 2 and 3. But for choice 4, how else would you train faster without better hardware? Unless you purposely used for loops where you shouldn’t have? In other words, you’re making too many assumptions, which I think is your problem with Q8 as well.
For Q8, whoever wrote the question and defined the correct answer apparently does not agree with you. Here again, I’m not looking at the quiz right now, so I have to take your word for it, but my guess is just that you’re just pushing the logic a bit too far about choosing “uniquely relevant” aspects of it.
Hi, I did include the question in my inquiry – I should have formatted it a bit.
Q3 - Recall the diagram of iterating over different ML ideas. Which of the stages shown in the diagram was improved with the use of a better GPU/CPU?
Q 8 - Why can an RNN (Recurrent Neural Network) be used to create English captions to French movies? Choose all that apply.
For Q3 - As is mentioned in the lecture, current driver for faster training is via Algorithmic Innovation. As Prof. Ng mentioned there are three drivers for performance scaleup - Data, Compute, and Algorithm. For algorithm he mentions use of ReLU instead of Sigmoid thus greatly reducing the gradient descent time at the two ends of sigmoid.
I would agree about my emphasis on “uniquely” – that is how I read the question - for Q3 I read what specifically is improved by GPU/CPU. And for Q8 I read it as what specifically makes RNN’s appropriate for creating captions.
Select all that apply often means that options are “ORs” - A or B or C – all may be true. In Q8 - being supervised alone would not do the trick for generating captions without RNN design.
However, RNN’s can be used for both supervised and unsupervised (per Google search). So I guess the right answer would be an “AND” and not inependent multiple choice questions.
Thanks.
Yes, I saw the questions, but you are making assertions about some of the answers and I can’t see all of them from what you wrote.
Was Q3 also a “check all that apply” question? Note that even though they may not say that, you can tell by whether they offer round or square checkboxes. Rounds ones for one correct answer and square for “choose all that apply”.
On Q8 as I said before: I think you are reading too much into it. I do not see the “uniquely” there.
Ok, I went and actually tried that Quiz again.
Q3 is a “choose all that apply” question, but in two tries, I had a different set of answers to choose from each time. And I agree that some of them were a bit ambiguous. So your mileage may vary.
Q8 clearly does not intend it to be exclusive or unique to RNNs. There is one option about the fact that it’s a supervised learning technique, which needs to be selected in order to get full points.