This paper studies application of Self-Organizing Maps (SOM) and Neural Gas (NGA). In chapter 2, we introduce growing structure to the NGA and SOM. The growing NGA is applied to skeltonization of the basic technology of image recognition. And we compare NGA to Growing SOM (GSOM). As image data input is given, the map topology can vary to adapt the input. If parameter are selected suitably. The skelton can be extracted efficiently. In chapter 3, we consider relationship between winner cell and neighbor cell the SOM. In update of neighbor cells of the winner, the learning coefficient changes depending on the past history of distance between cells. We apply the GSOM to Traveling Sales person Problems (TSP). We can find suitable tour with relatively small variation. In chapter 4, we consider parallel algorithm. We apply SOM to parallel algorithm. Performing parallel processing, we are indicated that execution time is considerably reduced. We compare parallel algorithm by SOM and to that by Radial Basis Adaptive Resonance Theory maps.

114 工学部 (旧) ・工学研究科 ＞ 学位論文 ＞ 修士論文