
Components of an artificial neuron
Inputs, x_{i:}
Typically, the input values are external stimulii from the
environment or come from the outputs of other artificial neurons.
They can be discrete values from a set, such as {0,1}, or
realvalued numbers.
[Back to the Artificial Neuron
applet page ]
Weights, w_{i}:
The first thing an artificial neuron does is to compute the
weighted sum of its inputs (i.e., the inner product between
the input pattern and the connection strengths). The weights are
realvalued numbers that determine the contribution of each input.
The goal of neural network training algorithms is to determine
the "best" possible set of weight values for the problem under
consideration. Finding the optimal set is often a tradeoff between
computation time, minimizing the network error, and maintaining the
network's ability to generalize.
[Back to the Artificial Neuron
applet page ]
Threshold, u:
The threshold is a real number that is subtracted from the weighted
sum of the input values. Sometimes the threshold is called a
bias value. In this case, the real number is added to the
weighted sum. For simplicity, the theshold can be regarded as
another input / weight pair, where
x_{0} = 1.
[Back to the Artificial Neuron
applet page ]
Activation Function, f:
The activation function for the original McCullochPitts neuron was
the unit step function. However, the artificial neuron model has
since been expanded to include other functions such as the sigmoid,
piecewise linear, and Gaussian.
The identity function is the simplest possible activation
function; the resulting unit is called a linear
associator.
The activation functions available in this applet are shown
in Table 1.
Unit Step 


Sigmoid 


Piecewise Linear 


Gaussian 


Identity


f (x) = x 
Table 1: Activation
Functions
[Back to the Artificial Neuron applet
page ]
Neuron Output,
y:
The artificial neuron computes its output according to the equation
shown below. This is the result of applying the
activation function to the
weighted sum of the
inputs, less the
threshold. This value can be discrete or
real depending on the activation function used.
Once the output has been calculated, it can be passed to another
neuron (or group of neurons) or sampled by the external
environment. The interpretation of the neuron output
depends upon the problem under consideration. For example, in
pattern classification, an output of 1 might imply the input
belongs to a certain class, whereas an output of 0 might mean that
it doesn't.
Back to the Artificial Neuron
applet page
