This animation demonstrates how a single hidden layer neural network using ReLU activation functions can approximate a curve. Each colored line is a single ReLU "hinge". The thick blue line is the network output (the sum of all hinges).