Approximate message passing (AMP) is a signal recovery algorithm for models involving large random matrices. An attractive feature of AMP is that its dynamical behavior could be characterized precisely via a simple iteration called state evolution (SE), in the high dimensional limit. However, the SE technique of AMP only works for matrices with i.i.d. Gaussian distributed elements. Orthogonal AMP generalizes AMP in the sense that it admits state-evolution characterization for a larger class of random matrices. [Learn more]

Estimation in Spiked Matrix Models