From April 19 to April 23, 2021, the TAMIDS Scientific Machine Learning Lab organized a one-week-long Hackathon to explore potential applications of graph learning in materials design. Our team consisted of two faculty members from the Department of Materials Science and Engineering, one research scientist from TAMIDS, and 8 graduate students drawn from the Department of Materials Science and Engineering, Department of Chemistry, and the Department of Electrical and Computer Engineering. Together, we were able to modify and deploy the MatErials Graph Network (MEGNet) model, one of the best libraries of graph learning in the material science field, on Grace, the new supercomputer at Texas A&M.
MEGNet has been widely used to automatically encode thousands of materials into crystal graphs, which are then used to train a graph neural net (GNN) to predict a material’s properties. Our team built off of MEGNet to create a GNN that would take in our own customized graphs, which we think may encode unique physics that may describe long-range interactions better than conventional crystal graphs. Additionally, we wrote programs to encode crystals into graphs using Python and to mine the Material Project database for building up our training data of known materials. We also managed to set up the software environment and train the model on the supercomputer Grace.
“As a graduate student pursuing a Ph.D. in material science, I have been very interested in applying graph theory to my own research”, said student participant Daniel Willhelm. “This Hackathon has provided a crash course on the tools I need, as well as deepened my fundamental understanding of graph learning. This was possible through the close collaboration and discussions with other students and Dr. Tao”.
|Jian Tao||TAMIDS/HPRC||Research Scientist||HPC, ML|
|Raymundo Arroyave||MSEN||Professor||ML, Material Discovery|
|Xiaofeng Qian||MSEN||Assistant Professor||ML, Material Discovery|
|Allison Arabelo||MSEN||Ph. D. Student||Material discovery|
|Cheng-Han Li||Chemistry||Ph. D. Student||ML, Material Discovery|
|Daniel Sauceda||MSEN||Ph. D. Student||Material Discovery|
|Guillermo Vazquez Tovar||MSEN||Ph. D. Student||Material Discovery|
|Brent Vela||MSEN||Ph. D. Student||Material Discovery, ML|
|Daniel Willhelm||MSEN||Ph. D. Student||Material Discovery|
|Nathan Wilson||MSEN||Ph. D. Student||Material Discovery|
|Ziyu Xiang||ECEN||Ph. D. Student||GNN|
Sponsorship and Awards
The organizers would like to acknowledge the generous support from Chevron, which provided awards for all student participants. The Hackathon MVP Award was shared by Daniel Sauceda, Daniel Willhelm, Nathan Wilson, and Ziyu Xiang.
Daily Schedule (April 19 – 23)
|9:00AM – 9:30AM||Kickstart the hackathon (on April 19) / Review the progress and discuss the plan for the day (other days)|
|9:30AM – 12:00PM||Morning session|
|12:00PM – 1:30PM||Lunch break|
|1:00PM – 3:30PM||Afternoon session|
|3:30PM – 4:00PM||Progress Report|