The authors show that the two-spiral problem can be easily solved usin
g a standard back propagation neural network by properly encoding the
raw input data. The authors also examine and compare several data enco
ding schemes for use in neural networks to improve the training effici
ency and generalisation property.