Rate-OptimalSubspaceEstimationonRandom Graphs
–Neural Information Processing Systems
Now it is sufficient to show thatσr0+1(Are) . This completes the proof forkˆM Mkop . The proof is an application of Fano's inequality. By Corollary 2, the right hand side is upper bounded by2kˆM Mkop/δ. The proof for r = 1 is simpler but the construction is slightly different.
Neural Information Processing Systems
Feb-10-2026, 13:29:02 GMT