Bits and Atoms - Exploring the Intersection of Machine Learning and Microscopy

Explore the intersection of machine learning and microscopy in this 48-minute lecture by Maxim Ziatdinov from Oak Ridge National Laboratory. Delve into the applications of AI in physics research, focusing on active learning models that interact with physical systems. Discover how microscopy serves as an ideal platform for materials discovery, physical learning, and controlled interventions. Learn about recent advancements in automated electron microscopy experiments, including object detection, atomic fabrication, and physics discovery through active learning. Examine the challenges of out-of-distribution drift in streaming image analysis and explore solutions. Investigate the limitations of simple Gaussian processes in active learning for complex systems and understand how deep kernel learning and structured Gaussian processes can enhance automated experiments for scientific discovery. Gain insights into the high-performance computing and edge infrastructure requirements for transforming modern microscopes into autonomous platforms for scientific breakthroughs.

Intro
About the Lab
Microscopes
Artificial Molecules
spectroscopic models
electron microscopy
previous work
sharing data
Variational models
Selfdriving microscope
Three types of experiments
Moving one atom at a time
Forward experiment
In principle
Problem with free trade neural nets
How to deal with uncertainty
Single vacancy lines
Inverse experiment
Optimization workflow
Deep Kernel Learning
Active Learning Invasion Optimization
Active Reports
Germani Mean Functions
Experiments
Selfdriving cars