Enabling Dynamic 3D Coherent Diffraction Imaging via Adaptive Latent Space Tuning of Generative Autoencoders

Coherent diffraction imaging (CDI) is an advanced non-destructive 3D X-ray imaging technique for measuring a sample's electron density. The main challenge of CDI is a lack of phase information in the intensity-based diffraction measurements, resulting in lengthy iterative CDI reconstruction processes that can return non-unique solutions. Furthermore, the CDI measurement technique is slow, requiring up to one hour for a single high resolution sample state measurement. The non-uniqueness of CDI reconstructions and the lengthy measurement process pose challenges for experiments that attempt to track the dynamic evolution of a sample through multiple states. We have developed an adaptive generative autoencoder approach for uniquely tracking a sample's electron density as it dynamically evolves. Our approach involves applying advanced adaptive control theoretic techniques directly within a low-dimensional latent embedding of a generative auotencoder, enabling a computationally efficient manner to account for time-varying shifting distributions in real-time. We analytically prove the convergence properties of our approach to the unique optimal solution for CDI over a wide class of functions. We also numerically demonstrate the method's ability to track sample evolution uniquely and to handle fast sample measurements with low signal to noise ratios with simulation-based synthetic data studies.