HowManySamplesisaGoodInitialPointWorthin Low-rankMatrixRecovery?

As a consequence, these global guarantees tend to be pessimistic, because the number of samples must be sufficiently large to eliminate spurious local minima everywhere, even at adversarial locations.