Backpropagation. Figure 1

Two networks are shown. Input units are shown as simple squares at the bottom of each figure. Other units are computational (designated by a horizontal line). Left: A standard 3-layer network. Four input units project to five hidden units, which in turn project to a single output unit. Not all connections are shown. Such a network is commonly used for classification tasks. Right: An example of a feed-forward network with four inputs, three hidden units, and two outputs