Sequential EM learning for subspace analysis

Subspace analysis is one of popular multivariate data analysis methods, which has been widely used in pattern recognition. Typically data space belongs to very high dimension, but only a few principal components need to be extracted. In this paper, we present a fast sequential algorithm which behaves like expectation maximization (EM), for subspace analysis or tracking. In addition we also present a slight modification of the subspace algorithm by employing a rectifier, that is quite useful in handling nonnegative data (for example, images), which leads to rectified subspace analysis. The useful behavior of our proposed algorithms are confirmed through numerical experimental results with toy data and dynamic PET images. [Copyright &y& Elsevier]