The other is the "negative" phase where the network is allowed to run freely, i.e. no units have their state determined by external data. Learning stops when statistics match in both phases. Statistics could be