Topic: Title: Independent component analysis for matrix-valued data Authors:  Joni Virta - University of Turku (Finland) [presenting]
Bing Li - The Pennsylvania State University (United States)
Klaus Nordhausen - University of Jyvaskyla (Finland)
Hannu Oja - University of Turku (Finland)
Abstract: In preprocessing high-dimensional matrix-valued data, e.g. images, a common procedure is to vectorize the observed matrices and subject the resulting vectors to one of the many methods used for independent component analysis (ICA). However, the structure of the original matrix is lost in the vectorization along with any meaningful interpretations of the rows and columns. To provide a more suitable alternative, we propose the Folded fourth order blind identification (FOLD), a matrix-valued analogy of the classic Fourth order blind identification (FOBI). In FOLD, instead of vectorizing, we stay in the matrix form and in a sense perform FOBI simultaneously on both the rows and the columns of the observed matrices. Furthermore, being an extension of FOBI, FOLD shares with it its computational simplicity. A simulated example is used to showcase the method's usefulness in discriminant analysis.