Phase Retrieval

Phase retrieval is the challenging task of reconstructing a signal from the magnitudes of its measurements, losing crucial phase information. Current research heavily emphasizes deep learning approaches, employing architectures like U-Nets, diffusion models, and coordinate-based neural networks, often incorporating physics-driven constraints or implicit generative priors to improve accuracy and robustness, particularly in noisy or low-light conditions. These advancements are significantly impacting various fields, enabling improved imaging techniques in microscopy, X-ray crystallography, and other areas where phase information is critical for accurate reconstruction. The development of faster, more sample-efficient algorithms remains a key focus, alongside addressing challenges posed by noise, limited data, and model misspecification.