The simplest neural network is a collection of nodes that have a
number of inputs from other nodes and maintain a number of outputs
in a binary on/off state.
Depending on the values or weights of the inputs,
the node/neuron will modify its output in accordance with its transition function.
Neural Nets perform a kind of
as the brain does
but in a vastly simplified manner.
In the early years of neural net research going back
to 1943 when W.S. McCulloch and W. Pitts published their paper
"A logical calculus of the ideas imminent in nervous activity"
Bulletin of Mathematical Biophysics.
Later, in the late 1960's the anticipation was that with a
large enough net we could simulate the human brain and obtain a
bottom up sort of emergent
This idea was short lived since the network would have to have near
100 billion nodes with over one million billion connections - like the human brain.
Until we can achieve this we can still solve
pattern matching and identification problems
that map very well to neural network systems.
These systems are the best choice for problems where
"Human Like" perception is required.
A lot of Neural Networks are now evolved using
the training algorithms.
Some of the problems associated with analysis of Neural Networks (NN) are the opaqueness of their internal workings.