How did we come to the conclusion that dw1 = x1*dz, dw2 = x2*dz and db = dz. This is mentioned in Week 2 content for gradient descent?

3 Likes

Hi @rohan.j,

The derivation is left as an exercise for the math interested student. You can try for yourself, by doing backprop on the following network:

2 Likes

@jonaslalin , thank you for your reply. I somehow happened to miss out the backpropagation/derivative of z wrt w1 which results in x1 multiplied by dZ which is similar for w2 and b as well.

2 Likes

Yes Using the chain rule:

```
dL/dw1 = dL/dz * dz/dw1 = dL/dz * x1
```

5 Likes

Some how the expression dw1 = x1 * dz looks incorrect.

since z = w1x1 + w2x2 + b

then dz/dw1 = x1

=> dz = x1 dw1 and not the other way around.

Please help me correct my understanding

In this course, dL/dw1 denotes as dw1.