DLS C5 W3 quiz has three errors, solutions proposed

The quiz for DLS C5 W3 has three errors. The goals of this post are (1) alert course staff to the errors so they may fix them and or (2) to allow learners to avoid unnecessarily failing the quiz - or perhaps worse, passing it when they shouldn’t.

Q2: In beam search, if you decrease the beam width B, which of the following would you expect to be true? Select all that apply.

One of the answers is “Beam search will converge after fewer steps.” In order for this option to be correct, it should read “Beam search will reach a solution after fewer algorithmic steps.” As it currently stands, this answer mischaracterizes beam search by implying that it takes steps until convergence, which it does not. Quoting an online resource, which comports with my understanding: “In beam search, convergence is not well-defined in the same way that it is for some other optimization algorithms, such as gradient descent or simulated annealing. This is because beam search is a heuristic algorithm that does not guarantee that the search will converge to an optimal or even feasible solution.”

Q4: Which of the following statements about alpha^<t,t’> are true? Check all that apply.

One of the answer options says “We expect alpha^<t,t’> to generally be larger for values of alpha^<t’>…” when it should be “We expect alpha^<t,t’> to generally be larger for values a^<t’>…” (emphasis added). Please change “alpha” to “a.”

Q10: In trigger word detection, if the target label for x^< t> is 1:

There are at least two versions of this question (a true / false one and a multiple choice one). All four of the answers for the multiple choice version of this question that I saw are wrong. There are two solutions here. One is to replace one of the wrong answers with a correct one like “x < t> represents the features of the audio at time t.” Another is to edit the quiz to use only the True/False version of this question.

Thanks for your message, I will notify the course staff.

Update: based on lab 2 (not the lecture) it appears that one of the answers to the Q10 quiz question might be consistent with lab 2 (if not the lecture). Note this excerpt from lab 2:

For each of the 1375 time steps, the model predicts whether someone recently finished saying the trigger word “activate”.

Since the quiz comes before the lab in the course order, the edits proposed above may be more learner friendly.