A fast learning algorithm for deep belief nets

It shows howtouse“complementarypriors”toeliminatetheexplaining- awayeffectsthatmakeinferencedifficultindenselyconnectedbeliefnets that have many hidden layers.