Abstract: Ptychography is a powerful computational technique in microscopy at all wavelengths and has enabled many applications, from semiconductors to biological specimens. In practice, obtaining accurate reconstructions requires simultaneously optimizing multiple parameters that are often selected based on trial-and-error, reducing the overall throughput, and even introducing human biases.
In this talk, I will discuss an automatic parameter tuning framework based on Bayesian optimization (BO) with Gaussian processes. With minimal prior knowledge, the workflow can produce high-quality ptychographic reconstructions that are superior to the ones processed by experienced experts. We also extend BO to other applications, such as experimental designs and tomographic reconstruction.