# how does gradient descent update weigths in neural network

Im currently trying to learn about back propagation, and it's going forward, but theres one thing that keeps me scratching my head, and doesnt really seems to be answered in any of the videos or articles im looking at. I understand now, that based on my loss, the weigths of my network is updated. But what i dont understand is how this happens. lets say i have this exercise network with the following weigths:

W_1 = 1.2 - w_2 = 0.4 - W_3 = 1.0


Now i do some training, and lets say i have the loss o.8. Now when i use my loss to update my weights, what happens specifically to the weights? are something being added, subtracted maybe multiplied?

Thanks a lot

• Gradient Descent takes the derivative of the loss with respect to the weights, then uses that value to update the weights. In your example you haven't given us the formula to calculate the loss, saying "the loss is 0.8" is insufficient information to perform the gradient descent step. A formula for the loss is needed. May 17, 2020 at 4:57

Here is the algo from tom mitchell's, in your case it shall be W_1 = w_1+delta,W_2 = w_2+delta, W_3 = w_3+delta.