I would like to know if the automatic differentiation mechanism in PyTorch will work if I use error backpropagation to train the final solution.