Minor error in youtube video[6:15:40] (Gradients of loss or parameters)

[himani007]

Hey,
I wanted to confirm if i am understanding this correct. It was mentioned in the youtube video on the time: 6:15:40 that Back propagation(4. loss backward) is moving backwards through the network to calculate the gradients if the parameters of our model w.r.t the loss.

But as i crosschecked and always understood was that we calculate the gradients of the loss w.r.t to parameters and not visa versa.

Please clear if this is a different thing you are referring to or am i right?

[apurvkumar023]

Well well welll I did read the maths behind it. what we actually do is
new_weight = old_weight - learning_rate*derivative of mean square error with respect to weight

and hat's how we get new weight.

[apurvkumar023]

and there is one more thing in which we sum the error to a node and then we try to minimise that but happens in big neural netwroks.

[himani007]

Well, thats wht i am sayijng, we take the derivative of the loss and not the parameters, but its said opposite in the youtube video. Must be a mistake!