Figure 2:
Auto-encoder with three layers.
