Entropy Spectral And Information On Clustering

Data clustering is a widely work in many real applications such as images and documents. In recent years, based on spectral method and information entropy,the research on clustering has attracted more and more interest and developed to produce two kinds of spectral method,namely spectral clustering and entropy clustering.Spectral clustering has properties of simplicity and high performance,and its performance is mainly affected by affinity matrix constructed by data. At present,neighbor propagation has some applications in constructing affinity ma-trix,and it includes Gauss-Seidel pattern and Jacobi pattern. At first,we in-troduce these two patterns, and then sketch the Gauss-Seidel pattern. The experiment results show that the affinity matrix constructed by our modified Gauss-Seidel pattern achieves better clustering results on spectral clustering model than the Gauss-Seidel pattern and the Jacobi pattern.Entropy clustering is constructing a corresponding model via analyzing the information entropy of data. Recently,a new model which is maximizing a sub-modular function subject to a matroid constraint achieves some success in data clustering. To solve this model, the improved greedy algorithm is applied. Due to the low efficiency of its computation, we make some improvement based on this algorithm, and propose an acceleration greedy algorithm. In addition,we give a new corresponding adaptive parameter selection of this model based on different datasets. The experiment results show that our acceleration greedy al-gorithm performs faster than the improved greedy algorithm, and the clustering performance is improved by using our new parameter section.