Deep Learning for Beginners

Figure 19 shows the visualization and the structure of the network. The network presents a
bias for each layer, always equal to +1, but with unknown weights. The inputs x1 and x2 are both
linked to the two neurons (i.e., NAND and OR) and these are linked together through an AND
function (in purple the True targets, the False targets are respectively in light-blue, in red, and in
orange).

Figure 19. The XOR framework.

Even if figure 19 can seem complex, in reality, we already know the majority of parameters.
We can start from the first layer, solving the NAND and the OR neurons (figure 20).

Figure 20. Parameters of the first layer.

Inputs and targets for XOR problem are:
x1 x2 y
0
0
0
0
1
1
1
0
1
1
1
0
and the activation function is always the hard-limit function.
29