Broader Impact: Since our method improves subspace clustering, it advances learning from unannotated data