The Perceptron
- The perceptron is a simple function that takes several inputs,
sums them, and passes them through another function to give a
single result.
- Summing just goes through the inputs and sums them.
- So, if the inputs are 3,5,-2 and 8, the sum 14.
- The function that the sum is passed through is called the transfer
functions.
- A simple function is a linear function, for example multiply by
2 or -0.3.
- If we take our sum, and pass it through the line we get 28 or -4.2.
- Another transfer function is the step function. If the sum is
greater than the threshold, 1 is output, if not 0 is.
- So if the threshold is 7, our sum gives 1. If the threshold is 42,
it gives 0. This is a lot like a McCullouch Pitts neuron with 1 being
a spike, and 0 no spike. This is why MLPs are called (incorrectly
in my view) neural nets.
- A third popular transfer function is the sigmoid function.
-
![Sigmoid Function](sigmoid.gif)
- There are lots of other transfer functions.