On Soft Learning Vector Quantization Based on Reformulation

The complex admissibility conditions for reformulated function in Karayiannis model is obtained based on the three axioms of radial basis function neural network. In this paper, we present an easier understandable assumption about vector quantization and radial basis function neural network. Under this assumption, we have obtained a simple but equivalent criterion for admissible reformulation function in Karayiannis model. We have also discovered that Karayiannis model for vector quantization has a trivial fixed point. Such results are useful for developing new vector quantization algorithms.