A doubt on backprop

Dear Mentor,

Could you please guide me the following issue?

There is a statement from the section 5.1.2 of the assignment “Convolution_model_Step_by_Step_v1”

Supposed that the dw should be in matrix form like this

May i know why it can be added up as shown in the statement quoted from the assignment?

Thank you.

I just checked the assignment, I can’t find this statement you mentioned in the assignment:
“dW = dW_11 + dW_12 + dW_21 + dW22”

In your example, dW will be a 2x2 matrix , and but the statement above assumes it is a scalar value (by adding 4 scalar values together). dW should be a matrix the same size as W, and we will eventually update, element-wise, each value of W based on dW.

Dear Mr Hackyon,

Could you please guide me on the reason behind this sentence?

Thank you.

Assuming dW is a 2x2 matrix, then dW_11 would be a single value from that 2x2 matrix. A single value without any dimensions is a scalar value.

For example, if you have a 2x2 matrix

A =
[[ 5 2 ],
[ 3 4 ]]

then A_11 would be 5. The 5 here is the scalar value. If you A_11 + A_12 + A_21 + A_22 = 5 + 2 + 3 +4 = 14, you will end up with another scalar value 14.