- on Education & Information
- Published in issue 3/2003 on page 0
Practical Neural Networks (3)
Feedback Nets and Competitive Nets
This month we look at two more advanced neural nets. The Hopfield network, which uses feedback in its structure, and the Competitive net which can recognise patterns in data, even if the programmer doesn’t know they exist.In 1983 a physicist named John Hopfield published a famous paper on neural nets. This paper helped to re-ignite the field, which had been languishing in the doldrums for some time.Actually, the ANN which bears his name — the Hopfield Network — is of somewhat limited practical use, but it does help us to understand the ins and outs of neural net behaviour.What Hopfield did was to create a network with feedback — with connections from the outputs, back towards the inputs. Figure 1 shows the idea. The network always has a single layer and the same number of inputs as neurons.