Question: Apply an MLP network with two sigmoid hidden units and a single linear output unit trained with the standard backpropagation learning algorithm to solve the XOR problem.
- Test the network sensitivity to various values of α; that is, test the convergence time of the algorithm for different values of α. Tip: start with α = 0.001 and vary it until α = 0.5.
- For a fixed value of α draw the decision boundaries constructed for each hidden unit, and the decision boundary constructed by the complete network.
- Can you solve this problem with a single hidden unit? Justify your answer.