Explore advanced techniques for processing 4D scanning transmission electron microscopy (4D-STEM) data using machine learning and Bayesian optimization in this conference talk. Discover how hierarchical unsupervised learning workflows can cluster nanobeam electron diffraction patterns, revealing essential features in materials such as strain and ripples in 2D lateral heterojunctions and ferroelectric domains in SnSe samples. Learn about a streamlined data-processing workflow for electron ptychography that employs Bayesian optimization with Gaussian processes to automatically tune reconstruction parameters, significantly improving efficiency and producing high-quality reconstructions. Gain insights into the challenges and advancements in multi-dimensional microscopy, including applications in cryogenic electron microscopy (cryo-EM) and the importance of sample preparation.
Assisting 4D-STEM Data Processing by Machine Learning and Bayesian Optimization
Institute for Pure & Applied Mathematics (IPAM) via YouTube
Overview
Syllabus
Intro
Epitaxial 2D lateral heterostructure
Graded 2D lateral heterojunctions
Large scale strain mapping
Ripples release the strain
Typical Workflow for 4D Data Processing
Odd diffraction intensity
Rotation invariant unsupervised learning
Uncovering ferroelectric domains
Machine leaming for nano beam 4D-STEM
Automate parameter tuning for ptychography
Bayesian Optimization with Gaussian Process
Cryogenic Electron Microscopy (cryo-EM) Cryo-EM
The bottleneck is sample preparation
Taught by
Institute for Pure & Applied Mathematics (IPAM)
