
4
Chapter 1
Perceptrons and Linear Separability
The perceptron [Rosenblatt, 1958] was the first artificial neural net-
work. It was computationally feasible on the hardware of that time,
was based on biological models, and was capable of
learning.
Unfortu-
nately, it also suffered from a significant weakness: the inability to
learn to perform an important family of classification tasks. Since the
original perceptron has little practical use today, it will not be
presented in detail. However, some discussion of it can contribute to
understanding closely related networks that are immensely practical.
Rosenblatt proposed many variations of the perceptron ...