The Diagnostic of Independence in ICA

• Main Authors: Chih-Hsiung Chang, 張智雄
• Other Authors: Yu-Fen Huang
• Format: Others
• Language: en_US
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/69337337419129347941

碩士 === 國立中正大學 === 統計科學所 === 94 === In multivariate analysis, independent component analysis (ICA) is a method to find the independent components of a latent variable model. There are many ways to find the independent components. One simple way is to make use of the property of the central limit theorem to search for the maximum nonguassianity in those combinations of the latent variables as one of the independent components. In order to check if the independent components found by ICA are ``really' statistically independent, we appeal to the Cramer-von Mises distance. If the Cramer-von Mises distance between two random variables is zero, then the two random variables are independent. Hence in this thesis, we first use negentropy as a measure of nonguassianity and the Cramer-von Mises distance, respectively, to search for the independent components. Furthermore we make comparison between these two methods.