Ming Du Email & Phone Number

Lemont, Il, Us

Leveraging AI, Bayesian statistics, and large-scale computing, I develop algorithms for data analysis and autonomous experimentation in the domain of X-ray science. My past and ongoing works include:- Using single-shot phase retrieval networks and image registration techniques to predict ptychographic scan positions without any initial guess;- Developing HPC-compatible training frameworks and deploying them on ALCF Polaris with >100 GPUs;- Developing transformer-based neural networks for crystallography structure prediction and training them on over 15 million samples;- Developing adaptive sampling and smart scanning approaches for x-ray imaging and spectroscopy using AI and Bayesian optimization.

Milpitas, California, Us

Milpitas, California, Us

• Developed image processing algorithms for modeling the image formation process in KLA’s high-precision photomask inspection systems used for sub-3-nm processes. Implemented algorithm solutions in the C++ codebase.• Led the development and productization a deep learning algorithm for defect detection, achieving 60% reduction of false positives. Optimized inference throughput using techniques and tools including OpenVINO. • Using computer vision and numerical optimization techniques, developed image correction and artifact removal algorithms that reduced the false positive rate of the photomask inspection system by over 70%. • Directed an internship project on modeling the artifacts generated in the image formation process using deep learning.

Lemont, Il, Us

• Developed Adorym, a reconstruction framework capable of performing large-scale 2D/3D phase retrieval, image reconstruction, and parameter refinement for multiple x-ray imaging techniques, built on automatic differentiation of PyTorch. • Developed a generative neural network-based method using “deep image prior” to reconstruct clean images from artifact-corrupted images in x-ray multislice ptychography. • Developed a distributed algorithm for large 2D wavefield propagation on high performance computers (HPCs). • Led the proposal and research of an ASCR Leadership Computing Challenge (ALCC) project, which is a highly selective HPC resource award, on large-scale 3D reconstruction.

Evanston, Il, Us

• Developed Tomosaic, a Python software package for beyond-field-of-view x-ray micro-CT. Achieved 3D reconstruction of a whole mouse brain with 10^13 voxels from 5.8 TB of raw data, using supercomputers at Ar-gonne Leadership Computing Facility. • Developed an innovative algorithm for 3D phase retrieval under complicated imaging scenarios, utilizing the automatic differentiation capability of deep learning tools.• Conducted x-ray tomography and ptychography experiments at the Advanced Photon Source.• Involved in the commissioning of an automated tomography processing pipeline Automo/Ripple at beamline 32-ID of the Advanced Photon Source. • Built an improved theoretical model for dose estimation in the x-ray imaging of thick hydrated specimens.

Singapore, Sg

• Experimentally studied the pore size dependence of anodic aluminum oxide on voltage and electrolyte temperature.• Developed a tool for unsupervised data analysis.

Doctor Of Philosophy - Phd, Materials Science

Bachelor Of Engineering, Materials Science And Engineering