Self-organized Formation of Various Invariant-feature Filters in the Adaptive-subspace SOM

The Adaptive-Subspace SOM (ASSOM) is a modular neural-network architecture, the modules of which learn to identify input patterns subject to some simple transformations. The learning process is unsupervised, competitive, and related to that of the traditional SOM (Self-Organizing Map). Each neural module becomes adaptively speciic to some restricted class of transformations, and modules close to each other in the network become tuned to similar features in an orderly fashion. If diierent transformations exist in the input signals, diierent subsets of ASSOM units become tuned to these transformation classes.