Multi Layer Perceptrons

• Multi Layer Perceptrons (MLPs) learning with backward propagation of error (backprop) are a very popular learning algorithm.
• The perceptron is a simple element that takes inputs and converts them to a single output.
• A layer of these, with the same inputs, cannot do some tasks.
• If there are multiple layers, with outputs from the first layer becoming the inputs to the second layer, you can approximate most interesting functions.
• MLPs are universal approximators.
• Using the Backpropagation algorithm, MLPs can be used to learn functions from sample input output pairs.