extend the code i have already done in java which creates a back propagation network with 20 inputs, 1 neuron in hidden layer and one output neuron.
This code has to be extended to have an extra hidden layer. Then it has to be tested with 12 different activation functions to see how each of them performs on the input.

If you are able to do this i will provide the code, fully commented.

If you wish you could do it using any other program but in that case the entire network would have to be created from scratch.