Hello! I was working on the quiz from week 3 and I got confused by a question.

One of the problems in this question asks about the dimension of b^[1] which, to my understanding, should be (4,m) where m is the number of training examples, but this is not an available option.

The reason I think it should be (4,m) is because the notation didn’t specify we are working on a specific training examples. In a later step in our programing we will have to add W^[1].X + b^[1] to find Z^[1] and for this operation to be defined, W^[1].X must be the same dimension as b^[1], which in this case will dictate that b^[1] be of the dimension (4,m). Could you please point out what I am confused about? I appreciate the help, thanks!

Hi @mo3tasem98 maybe you can have a look on the video on Week 3 “Explanation for Vectorized Implementation”, there around minute 1.20 Andrew shows the concept for the W matrix. To resume that concept in few words, I think you are confusing the fact that we want, for efficiency, use a minibatch of inputs (with size m) with the size of the network parameters which are minibatch independent. Well for b counts the similar reasoning done in the video for W.

Thanks a lot that clarified my confusion point. Andrew talked about the concept of python broadcasting in the same video which I forgot completely. This resolves the issue.

Hello,

I am a bit confused by the feedback about question 1 of the week 3 Quiz “Shallow Neural Networks”, below reported:

It seems to me that weights in row k of matrix w[j] apply to the neuron k in layer j, while the feedback seems to suggest that such weights are stored along a column.

Am I wrong?

Please let me know, thanks,. best regards,

Stefano Castiglione