Speaker
Description
Coherent diffraction imaging (CDI) and ptychography often suffer from the ill-posed nature of phase retrieval, particularly under low-dose, sparse-sampling, or missing-data conditions. This work examines several computational strategies leveraging Artificial Intelligence (AI) to improve reconstruction robustness in these scenarios. First, we implement a modern automatic differentiation-based ptychography pipeline built on the framework of [1]; second, we investigate the integration of Plug-and-Play (PnP) priors [2] as regularisers [3] within conventional iterative algorithms [4]; and finally, we evaluate the use of Deep Priors for end-to-end reconstruction [5]. Preliminary results suggest that these AI-driven approaches may offer increased robustness in challenging imaging conditions, specifically regarding high noise levels, low-dose acquisitions, and sparse sampling. Additionally, we discuss the utility of uncertainty estimation [5] as a means to assess the reliability of the reconstructed images.
References
[1] Guzzi F. et al., Condens. Matter 6(4), 36, (2021) doi:10.3390/condmat6040036
[2] Kamilov. U. S. et al., IEEE Signal Processing Magazine, vol. 40, no. 1, pp. 85-97, (2023), doi: 10.1109/MSP.2022.3199595
[3] Gianoncelli A. et al., JINST 21 C05007, (2026), doi:10.1088/1748-0221/21/05/C05007
[4] Guzzi F. et al., PeerJ Computer Science 8:e1036, (2022) doi:/10.7717/peerj-cs.1036
[5] Guzzi F. et al., JINST 21 C01006, (2026), doi:10.1088/1748-0221/21/01/C01006