This question asks you to compare and contrast neural networks with support vector machines.
(a) Suppose that a given neural network does not fit a training set closely enough, even after backprop has been run to termination. How can the structure of the network be modified to fit the training set more accurately?