Grid topologies for the self-organizing map.

摘要: The original Self-Organizing Feature Map (SOFM) has been extended in many ways to suit different goals and application domains. However, the topologies of map lattice that we can found literature are nearly always square or, more rarely, hexagonal. In this paper study alternative grid topologies, which derived from geometrical theory tessellations. Experimental results presented for unsupervised clustering, color image segmentation classification tasks, show differences among statistically significant most cases, optimal topology depends on problem at hand. A theoretical interpretation these is also developed.