Sanger's unsupervised learning is simply a procedure for plain Hebbian learning with constrained weight vector growth. This learning procedure is known to perform principal component analysis. This differs from Oja's learning rule in that the principal components are extracted in order, with respect to the output unit ordering.
