Multiple Layers

• When you have multiple layers of perceptrons, you have a multi-layer perceptron (or MLP).
• It turns out that almost any function can be approximated to an aribtrary degree of precision with a two layer MLP (given enough perceptrons).
• There is work in three and more layers.
• This is not the same thing as deep belief nets, but is related.
• The layers work in sequence, but there is parallelism within a layer.
• So, the first layer gets inputs. These are all the same, but may be differently weighted.
• Each applies their transfer function (typically the same), to produce outputs.
• These then feed inputs to the second layer.
• Since the weights are just mutliplications, and the transfer functions generally simple, this whole system is really fast.