Doctoral thesis: Multi-dimensional computational imaging from diffraction intensity using deep neural networks

Wednesday, April 13
12:30 pm


Iksung Kang


Diffraction of light can be seen everywhere, from the light shining out from curtains to multiple colors reflected from the surface of a CD. This light phenomenon explains any change in the path of the light due to an obstacle and is particularly of significance as a proper treatment of the diffracted light allows seeing transparent (or pure-phase) objects, e.g. biological cells under visible-wavelength light or integrated circuits under X-rays. However, light detectors only measure the intensity of the diffracted light upon the arrival of photons to their detection panels, which makes their measurements incomplete. Thus, this thesis addresses the extraction of multi-dimensional phase information out of these incomplete diffraction intensities by properly inverting them using deep learning across many applications. The inversion typically begins with the definition of a physical forward model that relates a diffraction intensity to its associated phase object and involves the physics-informing process of physics priors to deep neural networks. In this thesis, two-dimensional wavefront aberrations are retrieved for high-contrast imaging of exoplanets using a deep residual neural network, and planar transparent objects are revealed by a recurrent neural network through a dynamic scattering medium, both in an end-to-end training fashion. Next, a multi-layered, three-dimensional glass phantom of integrated circuits is reconstructed by limited-angle visible-wavelength phase tomography using a dynamical machine learning framework. Furthermore, a deep neural network regularization is deployed for the reconstruction of actual integrated circuits from several far-field diffraction intensities under the ptychographic X-ray computed tomography geometry.


  • Date: Wednesday, April 13
  • Time: 12:30 pm
  • Location: 24-307
Additional Location Details:

Thesis Supervisor: Prof. George Barbastathis

To attend this defense via zoom, please contact the doctoral candidate at