Global convergence of gradient descent for phase retrieval

2024/10/13 by Théodore Fougereux, Fougereux, Théodore, Cédric Josz +3
Physics and Astronomy · Engineering · #Advanced X-ray Imaging Techniques #Hydrocarbon exploration and reservoir analysis #Nuclear Physics and Applications

paper · pdf · doi:10.48550/arxiv.2410.09990

Abstract

We propose a tensor-based criterion for benign landscape in phase retrieval and establish boundedness of gradient trajectories. This implies that gradient descent will converge to a global minimum for almost every initial point.

Related

• The nonsmooth landscape of phase retrieval
• On Gradient Descent Algorithm for Generalized Phase Retrieval Problem
• Landscape Analysis of Simultaneous Blind Deconvolution and Phase Retrieval via Structured Low-Rank Tensor Recovery
• A note on Douglas-Rachford, gradients, and phase retrieval
• The Local Landscape of Phase Retrieval Under Limited Samples