Hands-On Unsupervised Learning with Python

Sanger's network

The Sanger's network model was proposed by Sanger (in Sanger, T. D., Optimal Unsupervised Learning in a Single-Layer Linear Feedforward Neural Network, Neural Networks, 2, 1989), in order to extract the first k principal components of a dataset, X, in descending order, with an online procedure (conversely, a standard PCA is a batch process that requires the entire dataset). Even if there's an incremental algorithm based on a particular version of SVD, the main advantage of these neural models is their intrinsic ability to work with single samples without any loss of performance. Before showing the structure of the network, it's necessary to introduce a modification to Hebb's rule, called Oja's rule:

This rule was introduced ...