Abstract
In the field of self-organizing maps (SOMs), a distance measure is one of essential factors for developing the feature map. The Euclidean distance measure is the most popular one, but it is also known that alignment of any training patterns will be change depending on how to define the measure. In the preceding studies, whose primal objective is pattern generation by a trained SOM, a fragmentized distance measure is introduced to generate some distinct intermediate patterns. Even though only symmetrical patterns, i.e., emoticons, are used for training, some asymmetrical patterns are contained in the generated patterns. Then, as a next step, further considerations are carried out in this study. As a result, it is confirmed that i) a sub-map corresponding to each fragment is developed independently at first, ii) all sub-maps are integrated into one, and iii) it makes possible to contain a variety of patterns in the entire feature map.
