We study a new class of spectral algorithms for low-rank estimation based on a tunable nonlinear deformation of an observed matrix.