One powerful way to do this is through a routine called slow reveal graphs.
Abstract: We present a linear-time subspace clustering approach that combines sparse representations and bipartite graph modeling. The signals are modeled as drawn from a union of low-dimensional ...