The Self-Organizing Map Applying the "Survival of the Fittest Type" Learning Algorithm

The self-organizing map that Kohonen has proposed maps high-dimensional vector data to low-dimensional space by phase conservation. And, it generates the feature map that visually catches the similarity among data. In addition, the reference vector where the unit in a competitive layer of SOM is achieved can interpolate an intermediate vector of the input vector data. In the pattern recognition of the class label, SOM that adds the class label to the element of the pattern and learns is especially called to be the supervised SOM. We propose SOM based on the survival of the fittest type learning algorithm to solve the problem of the delay and the over-training. As a result, the learning of the survival of the fittest type becomes possible, a needless node is excluded, and the probability density function can be presumed by the optimal number of nodes.