The University of Tennessee hosted an Artificial Intelligence and Machine Learning Microscopy Hackathon focused on solving automation and machine learning challenges in the field of microscopy.
Held last December, the two-day event brought together worldwide experts and enthusiasts from microscopy and machine learning to develop innovative solutions that bridge these fields. The Hackathon provided students the opportunity to develop code that enhances automation for control and data collection using cutting-edge microscopy tools.
“Microscopes generate data on millisecond timescales and humans process it on second or slower timescales. This gap makes machine learning essential for controlling instruments,” said Sergei Kalinin, Weston Fulton Professor in the Department of Materials Science and Engineering. “At the University of Tennessee, we’re uniquely positioned with electron and scanning probe microscopes directly interfaced with a supercomputer, enabling real-time instrument control and hence broad range of ML-driven automated experiments. Effectively, we have opened the door for the ML/AI agents to operate on the atomic and nanoscales.”
The Hackathon offered students a chance to contribute to projects that could lead to publications and recognition beyond the event. They also competed to earn prizes donated by industry and education sponsors, including the Office of Naval Research, ThermoFisher, and the Microscopy Society of America Student Council.
The highlight of the Hackathon was the set of problems based on digital twins of the instruments developed by Rama Vasudevan of Oak Ridge National Laboratory and his team, and Postdoctoral Researcher Boris Slautin of University Duisburg-Essen. The students got access to the static data sets and digital twins of the microscopes that allowed building active learning workflows.
The event began with an introduction and team-building phase led by second year Ph.D. candidate, Utkarsh Pratiush. Students teamed with academics and industry specialists around the world to solve Python coding challenges. Then, 80 participants worked for 36 hours, ending with nearly 20 project submissions in total.
Judges for the Hackathon used criteria based on teamwork, coding quality, originality, and what projects were publication ready.
Projects that were awarded cash prizes were:
- First place: GANder: Ferroelastic–Ferroelectric Domains Observed by Image-to-Image Translation, by Ralph Bulanadi (University of Geneva, team lead), Kieran J Pang (Justus-Liebig-Universität Gießen) and Michelle Wang (Technical University of Denmark).
- Second Place: AutoScriptCopilot, by Xiangyu Yin (Argonne National Laboratory), Yi Jiang (Argonne National Laboratory), Yu-Tsun Shao (University of Southern California) and Benjamin Fein-Ashley (University of Southern California).
- Third Place: Structure Discovery through Image-to-Graph Machine Learning Model, by Lauri Kurki (Aalto University, team lead), Harshit Sethi (Aalto University) and Jie Huang (Aalto University).
- Microscopy Society of America Student Council Award: Reward based Segmentation: Phase Mapping of 2D Polycrystalline Pd-Se Phases, by Kamyar Barakati (University of Tennessee, team lead) and Aditya Raghavan (University of Tennessee)
Kalinin and MSE Professor Gerd Duscher intend to integrate the winning script into instruments at the IAMM Electron Microscopy Core to test the effectiveness of AI driven data collection. All submitted codes are available on the official Hackathon website for researchers to freely download and utilize to advance capabilities with their instruments.
“The Hackathon was a tremendous success leading to UT students being independently recognized by the Microscopy Society of America Student Council for their outstanding work,” Duscher. “Looking forward, we aim to scale up this initiative and host an even larger hackathon in the future.”
Contact
Rhiannon Potkey (865-974-0683, [email protected])