[Use of automatic derivation]
For the required variable x, set require_grad == True
For the final function y, run backpropagation y.backward ()
Finally you can view dy / dx = x.grad
How to calculate the derivative of the middle quantity? ?
【High Dimensional Derivation】
Assuming that f is a high-dimensional function n to m, its derivation can be regarded as a jaco matrix nxm
You can enter an m-dimensional dz / dy for back propagation
y.backward (dz / dy)
In this case, x.grad returns dz / dx
Why can't you ask for guidance repeatedly? What is the principle inside?
【Neural Networks】