Quick Learning for Batch-Learning Self-Organizing Map

We have developed the batch-learning algorithm for the self-organizing map (BLSOM, Batch-Learning Self-Organizing Map) where the learning does not depend on the input order, and applied it to some analyses based on the similarity [1, 2]. The size of the map must be increased, as the number of the input vector increases, and the learning time grows in proportion to the product of them. In this study, we propose the fast learning algorithm for the BLSOM based from the following two viewpoints. (1) The initial map is made based on the principal component analysis. (2) The input vector does not move on the map in before and after the updating of the weight vectors very much.