Robust principal component analysis

Principal component analysis (PCA) is a technique used to reduce the dimensionality of data. In particular, it may be used to reduce the noise component of a signal. However, traditional PCA techniques may themselves be sensitive to noise. Some robust techniques have been developed, but these tend not to work so well in high dimensional spaces. This paper discusses the robustness properties of a recent PCA algorithm, SPCA(1). It shows theoretically and experimentally that this algorithm is less sensitive to the presence of outliers.