Lab 12: Multi Layer Perceptrons
- A perceptron takes inputs, sums them, and produces an output
based on a transfer function.
- Make a perceptron with two inputs; use the pass through
transfer function. So, it just sums the inputs and outputs
the sum. (A linear transfer function with a slope of 1 and no
offset is a pass through function).
- Make weights for the inputs. Replace the sum with the weighted
sum and put the weighted sum through the pass through transfer
- Make a linear transfer function, and set the slope to two.
Try it with an offset too.
- Now make two other perceptrons, with the same inputs, but
different weights and slopes for the transfer function.
- Do you see how one perceptron can duplicate the behaviour of
- Make a step transfer function. This can be used for a linear
- Now make a different layer of three perceptrons.
- Make the output from the first layer the input to the second.
- Weight the connections.
- Can you make an Xor Categoriser? You may need three layers.
- You can try to make your own data that is not linearly separable,
and make the categoriser categorise all of the data correctly
by setting the weights correctly.