Fig. 1 shows the neuron model for the ANN’Considered in
this paper. The adjustable, multiplicative weights correspond
to biological synapses. For the purpose of analytical modeling,
it is often convenient to allow a positive weight to represent
an excitatory connection and a negative weight an inhibitory
connection. A weight of zero is used when no connection
between a pair of neurons is to be made. The input transmitted
to a neuron through these weights may come from other
neurons or from external sources.
The weighted inputs to a neuron are accumulated and then
passed to an activation function which determines the neuron’s response. Commonly, a continuously varying, sigmoidal
activation function is used to model the frequency modulated action potential output of a biological neuron. The output of
the model neuron ranges between limits, such as 0 and 1, that
are analogous to a biological neurons minimum and maximum
firing rates. When an artificial neuron’s output is 0, the model
neuron is said to be “off”(or in state 0); the neuron is said to
be “on” (or in state 1) if its output is 1.