Automatic feature extraction using N-dimension convexity concept in a novel neural network
Chia-Lun John Hu · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1999
As we published in the last few years, when the pattern vectors used in the training of a novel neural network satisfy a generalized N-dimension convexity property (or the novel PLI condition we derived), the neural network can learn these patterns very fast in a NONITERATIVE manner. The recognition of any UNTRAINED patterns by using this learned neural network can then reach OPTIMUM ROBUSTNESS if an automatic feature extraction scheme derived from the N-dimension geometry is used in the recognition mode. The simplified physical picture of the high-robustness reached by this novel system is the automatic extraction of the most distinguished parts in all the M training pattern vectors in the N-space such that the volume of the M-dimension parallelepiped spanned by these parts of the vectors reaches a maximum.