Question about DLS Course4 Week2 quiz


I got one quiz question wrong and don’t understand the reason.

It is question 7. “Making an inception network deeper won’t hurt the training set performance”. I believe there is a deep inception network shown in the slides but the comment says it is wrong. Also I don’t quit get the comment, saying “when stacking more layers the training performance might start increasing instead of decreasing”. That sounds exactly the same as “it won’t hurt the performance”.

Could anyone explain what does this mean?

Thank you.

My interpretation is that the point is that it is not a guarantee that adding more inception layers will help. It might increase the performance, but it might also decrease it. In other words the way the question is phrased, it is asserting that it always helps. That is false, because there’s essentially nothing in DL that you can say that about (that it is always guaranteed to help). Everything is situational: even if it helps in most cases, there will be some outlier cases where it doesn’t.