Figure 3: Auto-encoder with multiple hidden layers.