UKF Based Self-organizing Feature Maps Algorithm for Serial Analysis of Gene Expression Data

摘要

Due to the higher dimensional and nonlinear properties of the serial analysis of gene expression data, traditional self-organizing feature maps can't clustering effectively. To circumvent the parameters study of the self-organizing feature maps, a novel algorithm based on the Kalman filter and the unscented transform is presented. During the learning process, the learning coefficient and the width of the neighborhood function can updated automatically according to the input data. By clustering the mouse retinal SAGE data, results show that the novel algorithm has competence.