A simple neuron feature detection
Mohamed L. Hambaba · 2003
A novel approach is discussed relative to unsupervised learning in a single-layer linear neural network. An optimality principle is proposed which is based on preserving maximal information in the output units. The unsupervised learning rule is based on a Hebbian learning rule. This learning rule finds the principal components of the input correlation matrix. For patterns classification, and image coding, the authors have modified the learning rule which finds the Boolean eigenvector