Johnson-Lindenstrauss Property Implies Subspace Restricted Isometry Property
share
Dimensionality reduction is a popular approach to tackle high-dimensional data with low-dimensional nature. Subspace Restricted Isometry Property, a newly-proposed concept, has proved to be a useful tool in analyzing the effect of dimensionality reduction algorithms on subspaces. In this paper, we establish the subspace Restricted Isometry Property for random projections satisfying some specific concentration inequality, which is called by Johnson-Lindenstrauss property. Johnson-Lindenstrauss property is a very mild condition and is satisfied by numerous types of random matrices encountered in practice. Thus our result could extend the applicability of random projections in subspace-based machine learning algorithms including subspace clustering and allow for the usage of, for instance, Bernoulli matrices, partial Fourier matrices, and partial Hadamard matrices for random projections, which are easier to implement on hardware or are more efficient to compute.
READ FULL TEXT