Misunderstand

[thieu1995]

Hi Omar,
I read your article, I have some places which I don't understand. Could you please explain clearly for me?
I denote, d(2,3) mean matrix d with dimension (2-row, 3 col)
dh2/dz2 = h2*(1-h2). Why we can multiple vector here? h2(2,1); (1-h2) => t1 (2,1).
+) so h2(2,1) * t1(2,1) = ???

dh2/dz2 = h2(1-h2) = [0.37155696 0.454142644] (1- [0.377166 0.454142644]) = [-0.233353037 0.24789]
+) why the result has "minus" on [-0.2333..] ???

dLoss/dW1 = [1.766405 0.40015721]^T
what ^T mean here?

[omar-florez]

Hi Thiev,

1. That operation is possible because it is a matrix multiplication and not a dot product. Here more details: https://www.physicsforums.com/threads/matrix-multiplication-vs-dot-product.607749/
2. Yup, that's a typo. All values in h2/dz2 should be positive. Will fix the figure later, the final values of the resulting matrix should be fine.
3. That T just means the transpose of the matrix

[thieu1995]

tks for the answer. It's clear for me now.

[omar-florez]

Glad to know :)