Hebbian learning subspace method: A new approach

摘要

In this paper, we propose a new learning algorithm for the Subspace Pattern Recognition Method (SPRM) called the Hebbian Learning Subspace Method (HLSM). It uses the notion of a weighted squared orthogonal projection distance which gives different weightages to different basis vectors in the computation of the orthogonal projection distance. The principle applied during learning is the same as that used in the earlier Learning Subspace Method (LSM): the projection on the wrong subspace is always decreased and the one on the correct subspace is always increased. We also propose a neural implementation for the HLSM. Experiments have been conducted on an extensive numeric set of handprinted characters involving 16659 samples using the SPRM, the HLSM and the Averaged LSM. Excellent results have been obtained using all the subspace methods thus demonstrating the suitability of subspace methods for this application.