Watch data flow forward through a neural network, then gradients flow backward. The network learns by adjusting its weights to minimize the loss.