The Study And Application Of Model Cluster Based On EM Algorithm

There are many applications that require the parameter estimation of data model, such as artificial intelligence, pattern-recognition and machine-learning. It is often desired to estimate the maximum-likelihood or maximum-posterior likelihood. EM algorithm, which is named expectation maximum algorithm, is a general-purpose algorithm for maximum likelihood estimation in a wide variety of situations best described as incomplete-data problem. Its core idea is to literately compute the likelihood function until it converges to some optimal value for the given data, This paper introduced the basis of cluster in brief and reviewed the typical cluster methods, and focus on the. In the following of the paper, we further studied the EM algorithm from four aspects:1, Implemented the EM algorithm and compared with the kmeans algorithm on some application cluster. It also can be used as the basis of the further study.2, The performance of EM algorithm heavily depends on the initial values of the parameters. When EM algorithm is utilized to realize Gaussian-Mixture-Model based clustering, how to initialize it becomes a pivotal issue. In this paper, the binning method is adopted to initialize EM on the base of comparison of other methods. Our experimental results demonstrate that Gaussian-mixture-model based on clustering using EM with the binning method for initialization outperforms those with other classical initialization methods.3, Semi-supervised clustering employs a small amount of labeled data to aid clustering analysis. The EM algorithm based on dual Gaussian mixture model has been studied with the added labeled samples as the initial parameters in this paper. Our experimental results demonstrate that the algorithm increase the recognition rate for samples and has good clustering ability and some application fields.4, Finally, This paper focuses on the speaker recognition system based on Mel-frequency Cepstral coefficients and GMM. It also gives the theory basis and processing arithmetic to compute MFCC in detail;Then influnces on recognition performance of GMM mixure component and the number of EM iteration are discussed by the experiments.