XSCOPE/LANS Seminar-Two Speakers

Date/Time: August 28, 2024/ 11:00 AM-12:00 PM
Location: See Meeting URL on the cels-seminars website which will require an Argonne login.

Seminar Title: Refik Mert Cam Ptychographic Image Reconstruction Using Denoising Diffusion Models

Speaker: Refik Mert Cam, PhD Candidate, Department of Electrical and Computer Engineering, University of Illinois Urbana-Champaign (UIUC)

Description: X-ray ptychography is a powerful imaging technique used in materials science, biology, and nanotechnology for achieving high-resolution images. It involves scanning a coherent X-ray probe across a sample in overlapping regions, with a detector capturing the diffraction patterns at each position. Conventional image reconstruction algorithms often require substantial overlap, leading to increased data volume and prolonged experiment times. To address these challenges, we propose a novel reconstruction method based on denoising diffusion probabilistic models. These models capture implicit object priors, allowing them to effectively regularize the ill-posed nature of the ptychographic inverse problem. Our simulations demonstrate that this approach can achieve high-quality reconstructions while significantly reducing data requirements, offering a more efficient alternative to traditional methods.

Bio: Refik Mert Cam is a PhD candidate in the Department of Electrical and Computer Engineering at the University of Illinois Urbana-Champaign (UIUC). He earned his B.S. in Electrical and Electronics Engineering from Middle East Technical University, Turkiye, in 2020. His research focuses on inverse problems, image reconstruction algorithms, and the application of deep learning and generative modeling in imaging. Refik has been recognized with the Elsa and Floyd Dunn Award from the UIUC ECE department and was named a 2023-2024 Mavis Future Faculty Fellow in UIUC Grainger Engineering.

Seminar Title: Canberk Ekmekci Bayesian Inversion for Ptychography with Deep Generative Priors

Speaker: Canberk Ekmekci,  Ph.D. Candidate, Department of Electrical and Computer Engineering, University of Rochester

Description: Ptychography, a cutting-edge scanning coherent diffraction imaging technique, enables the visualization of features smaller than 10 nanometers. Despite its potential, the method is often hindered by large data volumes and extended acquisition times. This talk will explore a novel approach to overcoming these challenges by leveraging Bayesian inversion techniques alongside deep generative models. By employing a pre-trained generative model to simulate ptychographic objects, this approach enhances reconstruction quality even under sparse data conditions and has the potential to optimize experimental design for more efficient data acquisition. Preliminary results from synthetic simulations indicate significant advancements in accelerating data acquisition and in quantifying the inherent uncertainties associated with the ptychographic inverse problem.

Bio: Canberk Ekmekci is a Ph.D. candidate in the Department of Electrical and Computer Engineering at the University of Rochester, studying under the supervision of Prof. Mujdat Cetin in the Signal, Data, and Imaging Sciences Lab. He earned his Bachelor of Science degree in Electrical and Electronics Engineering from Koç University in Istanbul, Turkey, in 2019, where he graduated as the salutatorian.His research focuses on computational imaging, with a particular emphasis on the development and analysis of algorithms for imaging inverse problems. His recent work includes learning-based modeling for imaging applications, Plug-and-Play algorithms, deep algorithmic unrolling for image reconstruction, automatic parameter tuning for iterative algorithms, and uncertainty characterization in imaging inverse problems. He has been awarded the Robert L. and Mary L. Sproull Fellowship by the University of Rochester.