Lets try to generalize a sine wave using neural network. The NN will learn the data and will approximate. We will need atleast 20 hidden neurons for the NN to model properly. For a NN to generalize to a non-linear output, it needs atleast 20 neurons in the hidden layer.
I used about 20 neurons in the hidden layer and trained for about 800 epocs below you see the result of the learning capability of the neural network.
Red – Original Training data
Blue – NN output after learning
