ABSTRACT
In this chapter, we discuss some popular methods for multivariate analysis and explore how they can be “sparsified”: that is, how the set of features can be reduced to a smaller set to yield more interpretable solutions. Many standard multivariate methods are derived from the singular value decomposition of an appropriate data matrix. Hence, one systematic approach to sparse multivariate analysis is through a sparse decomposition of the same data matrix. The penalized matrix decomposition of Section 7.6 is well-suited to this task, as it delivers sparse versions of the left and/or right singular vectors.
