A non convex minimization approach for phase retrieval


Peng-Wen Chen

12:30:00 - 14:30:00

103 , Mathematics Research Center Building (ori. New Math. Bldg.)

Phase retrieval is a problem regarding  recovering a signal (n-vector) from the magnitude of its Fourier transform, which  is a special case of recovering an n-vector from the magnitude of N phaseless linear measurements. PhaseLift is a convex formulation, which lifts the unknown signal to a n-by-n matrix with rank-one constraints. In  theory, when N\\ge 4n-2, the rank-one matrix is uniquely determined. In this talk, we will study a non convex approach based on augmented Lagrangian methods. Experiments demonstrate the effectiveness of this approach.