Neural network depth significantly affects training loss.Weights are initialized randomly using Gaussian noise.Performance metrics are computed using mean squared error.The direction of weights is adjusted opposite to the gradient to reduce loss.