While
there are likely many reasons for these mixed results, two reasons that are frequently discussed are
the inability of most optimization algorithms to obtain an optimal set of weights for the neural
network and the fact that it is common for neural networks to be over parameterized