Freiburg 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

A: Fachverband Atomphysik

A 10: Interaction with Strong or Short Laser Pulses I (joint session A/MO)

A 10.3: Vortrag

Dienstag, 12. März 2024, 11:45–12:00, HS 1010

Reconstruction of Three Dimensional Molecular Density from XFEL Scattering Images using Machine Learning — •Siddhartha Poddar, Ulf Saalmann, and Jan Michael Rost — Max Planck Institute for the Physics of Complex Systems

As the three-dimensional electron density profile recovery technique for a single macro-molecule from a large dataset of coherent diffraction images generated using an X-ray free-electron laser, I have applied an unsupervised machine learning algorithm namely Generative Adversarial Network (GAN). It learns to mimic the high-dimensional distribution of given images by generating its own 'fake' distribution of images with the help of a deep convolutional neural network called the discriminator which distinguishes samples drawn from the original and fake distributions. To generate samples for this fake distribution of images, GAN creates and constantly modifies a three-dimensional structure. This structure is claimed to be unique and an equivalent version of the target electronic density profile of the molecule.

Keywords: Machine Learning; XFEL imaging; 3D Reconstruction; Phase Retrieval; Image Processing