1
Artificial Neural Networks
There is nothing either good or bad but thinking makes it so.
—Shakespeare, Hamlet, II, ii
Since man’s earliest efforts to build an electronic calculating machine,
scientists and engineers have dreamed of constructing the ultimate arti-
ficial brain. Though we may never reach this goal, the first successful
attempt to create a computer algorithm that would mimic, albeit in a
much simplified way, the brain’s remarkably complicated structure and
function represented a significant stride forward. These algorithms,
known as artificial neural nets, are defined as an interconnected group
of information processing units whose functionality is roughly based on
the living neuron. As these units “learn” or process information by
adapting to a set of training patterns, it is reflected in the strength of
their connections.
Neural nets represent a different paradigm for computing than that of
conventional digital computers, because their architecture closely paral-
lels that of the brain. (Traditional computers, based on von Neumann’s
design, were inspired by a model of brain function by incorporating con-
cepts such as input, output, and memory, but reflect this only abstractly
in their architecture.) Neural nets are useful for problems where one
can’t find an algorithmic solution, but can find lots of examples of the
sought after behavior, or where we need to identify the solution’s
structure from existing data. In other words, they don’t need to be
programmed to solve a specific problem; they “learn” by example. They
have their roots in a pioneering 1943 paper written by mathematician
Walter Pitts and psychiatrist Warren McCullough, “A Logical Calculus
of the Ideas Immanent in Nervous Activity.” It was the first time anyone
1
Artificial Neural Networks
There is nothing either good or bad but thinking makes it so.
—Shakespeare, Hamlet, II, ii
Since man’s earliest efforts to build an electronic calculating machine,
scientists and engineers have dreamed of constructing the ultimate arti-
ficial brain. Though we may never reach this goal, the first successful
attempt to create a computer algorithm that would mimic, albeit in a
much simplified way, the brain’s remarkably complicated structure and
function represented a significant stride forward. These algorithms,
known as artificial neural nets, are defined as an interconnected group
of information processing units whose functionality is roughly based on
the living neuron. As these units “learn” or process information by
adapting to a set of training patterns, it is reflected in the strength of
their connections.
Neural nets represent a different paradigm for computing than that of
conventional digital computers, because their architecture closely paral-
lels that of the brain. (Traditional computers, based on von Neumann’s
design, were inspired by a model of brain function by incorporating con-
cepts such as input, output, and memory, but reflect this only abstractly
in their architecture.) Neural nets are useful for problems where one
can’t find an algorithmic solution, but can find lots of examples of the
sought after behavior, or where we need to identify the solution’s
structure from existing data. In other words, they don’t need to be
programmed to solve a specific problem; they “learn” by example. They
have their roots in a pioneering 1943 paper written by mathematician
Walter Pitts and psychiatrist Warren McCullough, “A Logical Calculus
of the Ideas Immanent in Nervous Activity.” It was the first time anyone
