Two phases

1. One node wins the competition

2. That node learns, others not

• Nodes start off with random weights

• No `correct' output connected with inputs: unsupervised learning