What is wrong with back-propagation?
		• It requires labeled training data. – Almost all data is unlabeled. • The learning time does not scale well – It is very slow in networks with multiple hidden layers. • It can get stuck in poor local optima.