Most recent results in matrix completion assume that the matrix under
consideration is low-rank or that the columns are in a union of low-rank
subspaces. In real-world settings, however, the linear structure underlying
these models is distorted by a (typically unknown) nonlinear transf