Title: Ptychography reconstruction: Fast optimization methods from a signal processing perspective

Abstract: Ptychography, a versatile computational imaging technique that jointly reconstructs the sample's complex function and the illumination probe from a series of overlapping intensity measurements, is a challenging nonconvex inverse problem whose applications span across physics, materials science, and biology.

From a signal processing standpoint, existing reconstruction methods can be broadly categorized according to their optimization paradigm, namely alternating optimization schemes—such as Error Reduction (ER) and extended Ptychographic Iterative Engine (ePIE)—and simultaneous optimization approaches, exemplified by Wirtinger Flow.

This talk revisits this classification and provides a unified analysis demonstrating that classical acceleration techniques, including Nesterov-type momentum, can be naturally incorporated. This perspective also offers new insight into the historical limitations of the ePIE algorithm family and explains why their convergence properties were not fully leveraged in earlier implementations.

Furthermore, in this talk I will also introduce an illumination probe positions' uncertainty aware AO-based algorithm whose practical RoC (rate of convergence) is either competitive or surpases the performance of several state-of-the-art Ptychography reconstruction algorithms.

Title: Ptycho-Gramian: the inner products between frames and correlated beam instabilities

Abstract: Ideally, in ptychography, each pair of exit waves is related by a spatial shift and a common illumination function. In practice, beams fluctuate and samples drift, adding up exit waves incoherently onto the detector. Here we investigate the pairwise discrepancy among reconstructed frames to scale up convergence rates for large scale problems. Joint work with Yuan Ni (SLAC) and Huibin Chang (Tianjin Normal University)

Bio: Master in Physics in Parma, Italy. Ph.D. in Physics, in Grenoble, France (2000), Lawrence Berkeleley National Lab (2000-2002, 2007-2020), Lawrence Livermore National Lab (2003-2007), SLAC/Stanford (2021-present). Directors awards for scientific achievements at LBNL, LLNL, SLAC, and Optica Fellow. 

Abstract: In this talk we discuss the use of generative models as priors for a robust reconstruction in ptychographic inversion with high uncertainty. As a simplified model towards ptychographic single particle imaging we consider the reconstruction in ptychography with unknown positions. We compare different approaches and discuss the potential and limitations of the approach.  

Bio: Martin Burger is a Lead Scientist at DESY and Professor of Mathematics at the University of Hamburg. He heads the Computional Imaging group at DESY, which is also a research unit of the Helmholtz Imaging cooperation platform. Together with his team, he develops imaging reconstruction methods and novel mathematical tools with the aim of improving the reconstruction and analysis of scientific images. Moreover, his group is concerned with the mathematical foundations of machine learning and of algorithms in data science.