The team of researchers was composed of Zico Kolter, Shaojie Bai, Devin Wilmott, Mordechai Kornbluth, and Jonathan Mailoa, who won a Kaggle competition onPredicting Molecular Properties this past September.

Shaojie Bai, a doctoral student with the Machine Learning Department and team member, said the team completed the project during an internship at the Bosch Center for AI (BCAI).

During the competition, the teams needed to develop machine learning algorithms that can predict the magnetic interaction between two atoms in a molecule (i.e., the scalar coupling constant).

Bai said current quantum mechanical methods can model the magnetic interactions between atoms in a molecule accurately, but using physical methods would take days or weeks to model a “single molecule.” He said the team used machine learning methods to make this prediction millions of times faster without sacrificing accuracy.

Ultimately, such tools will enable researchers to make progress in a range of significant problems, like designing molecules to carry out specific cellular tasks or designing better drug molecules to fight disease.

A more detailed writeup regarding this solution can be found onKaggle’s website, and the code can be found onGitHub.

For More Information