THE SOM METHOD WITH TUNABLE KERNEL FUNCTION

摘要

The Self-Organizing Maps is an unsupervised learning algorithm. This algorithm shows poor robust and reliability when the distribution of study samples has many states and becomes highly nonlinear. The study based on kernel function, performing a nonlinear data transformation into some high dimensional feature space, increases the probability of the linear reparability of the patterns within the feature space. But be aimed at the different data, the classification effect of various kernel functions is different. So the choice of kernel function is depend on the questions. In this paper, a tunable kernel function algorithm is proposed. By studying and adjusting modulus, the effect is better than the result by using single kernel function.