Poisson-Based Self-Organizing Neural Networks for Pattern Discovery

摘要

Based on unsupervised learning paradigm, self-organizing neural networks have achieved great success in applications of automatic pattern discovery. However, the development of self-organizing neural networks is traditionally based on the assumption that data are governed by a normal distribution. Application of self-organizing neural networks in the areas where data are better modelled by other statistical distributions such as a Poisson distribution has received less attention. Based on the incorporation of the statistical nature of data with a Poisson distribution into a Self-Organizing Map, this paper presents a Poisson-based self-organizing neural network. The proposed network has been tested on two datasets including a real biological example. The results indicate that, in comparison to traditional self-organizing maps, the proposed model offers substantial improvements in pattern discovery in data governed by a Poisson distribution.