Robust Bilinear Probabilistic Principal Component Analysis

Principal component analysis (PCA) is one of the most popular tools in multivariate exploratory data analysis. Its probabilistic version (PPCA) based on maximum likelihood procedure provides a manner to implement dimension reduction. Recently, bilinear PPCA (BPPCA) model, which assumes that noise terms follow matrix variate Gaussian distributions, has been introduced directly deal with two-dimensional (2-D) for preserving structure 2-D data, such as images, and avoiding curse dimensionality. However, distributions are not always available real-life applications may contain outliers within sets. In order make BPPCA robust outliers, this paper, we propose model under assumption t terms. The alternating expectation conditional maximization (AECM) algorithm used estimate parameters. Numerical examples several synthetic publicly sets presented demonstrate superiority our proposed feature extraction, classification outlier detection.