Low-rankness well expresses the low dimensional structure of high dimensional data. Instead of using PCA which is plagued by a well-known, well-documented sensitivity to outliers, recent methods minimize the trace-norm of data matrix.
RPCA:
Where D is the lowrank structure of X, and E is the error term. Using trace norm, it will robust to noise and outliers.
Face recognition:
Low-Rank Representation (LRR) is an effective method for exploring the multiple subspace structures of data. It use the data matrix as a dictionary.
Normal estimation