Implications of Self-organization: Building Vector Quantizers and Classifiers with Self-organizing Maps

Self-organizing maps (SOMs) are among the most interesting classes of neural networks due to their ability to map a non-linear, high-dimensional data space onto a lower dimension, regular lattice space. The resulting mapping exhibits two useful properties: topology preservation and density matching. This self-contained book discusses topology preservation and density matching properties of SOMs and their implications explicitly in the context of pattern recognition tasks. It also looks at how to exploit SOMs for improving system performances.