A NN recognizes the archetypes of 1 and 0
and correctly classifies them.
If the input is the archetype of 1,
the NN must output 10,
if it is that of 0,
the output must be 01.
If the input is any other,
the output must be 00.
The output 11 must never appear
because that amounts to a contradiction: it means
that a given bitmap represents both archetypes,
that of 1 and that of 0.
The code is prepared for noise tolerance:
an additional bias with arbitrary value
is declared for each perceptron.
Contradictions are expected
because value assignation is arbitrary.