Last answered:

04 Nov 2021

Posted on:

02 Apr 2021


Gradient of the loss function with respect to weights

Thank you very much for this great lecture. Although I understood the most of it, one question remained. What steps of derivation were taken to get from Wi+1 = Wi - n ∇W L(y,t)
to  Wi+1 = Wi - n ∑ Xiδi

If describing the entire process will be troublesome for you for some reason, could you please refer me to some online resources?

Thank you.

2 answers ( 0 marked as helpful)
Posted on:

27 Oct 2021


Hello, it is late but let me explain. You should check out the "partial derivative" concept. ∇W means that it depends on w. So when we take partial derivative of it, w is assumed as only variable and the others as constant. Therefore, you get: 1/2 * 2 * (xi * w + b)^(2-1) * xi ; when you take derivative of it.

Posted on:

04 Nov 2021


Great content, but I have the same doubt. I understand why the derivative is what it is, but I don't get where does the sum comes from. Thanks in advance

Submit an answer