自组织映射34部分.ppt

The Self-Organizing MapⅢ、Ⅳ Ⅲ. Fine Tunning of the Map by Learning VectorQuantization(LVQ) Methods If the Self-organizing Map is to be used as a pattern classifier in which the cells or their responses are grouped into subsets, each of which corresponds to a discrete class of patterns, then the problem becomes a decision process and must be handled differently. The original Map, like any classical Vector Quantization (VQ) method is mainly intended to approximate input signal values, or their probability density function, by quantized “codebook” vectors that are localized in the input space to minimize a quantization error functional . On the other hand, if the signal sets are to be classified into a finite number of categories, then several codebook vectors are usually made to represent each class, and their identity within the classes is no longer important. In fact, only decisions made at class borders count. It is then possible, as shown below, to define effective values for the codebook vectors such that they directly define near-optimal decision borders between the classes, even in the sense of classical Bayesian decision theory. Ⅲ. Fine Tunning of the Map by Learning VectorQuantization(LVQ) Methods Type One Learning Vector Quantization (LVQ1) Type Two Learning Vector Quantization (LVQ2) Type Three Learning Vector Quantization (LVQ3) Ⅲ. Fine Tunning of the Map by Learning VectorQuantization(LVQ) Methods Ⅲ. Fine Tunning of the Map by Learning VectorQuantization(LVQ) Methods A. Type One Learning Vector Quantization (LVQ1) If several codebook vectors are assigned to each class, and each of them is labeled with the corresponding class symbol, the class regions in the space are defined by simple nearest-neighbor comparison of with the ; the label of the closest defines the classification of . To define the optimal placement of in an iterative learning process, initial values for them must first be set using any classical VQ method or b