Sign In

Communications of the ACM

ACM TechNews

Machine Learning Program for Games Inspires Development of Groundbreaking Scientific Tool

Part of a graph showing excellent algorithm performance for force field predictions of elemental nanoclusters covering 54 elements in the periodic table.

“Our work represents a major step forward in this sort of model development for materials science,” said Subramanian Sankaranarayanan, group leader at Argonne’s Center for Nanoscale Materials and associate professor at the University of

Credit: Argonne National Laboratory

A reinforcement learning algorithm developed by researchers at the U.S. Department of Energy's Argonne National Laboratory aims to speed up materials discovery by modeling material properties at the atomic and molecular scale.

The researchers built decision trees into their algorithm to provide positive reinforcement based on the degree of success in optimizing model parameters.

Using the carbon cluster of computers in Argonne's Center for Nanoscale Materials, the Theta supercomputer at the Argonne Leadership Computing Facility, and computing resources at the National Energy Research Scientific Computing Center, the algorithm was tested with 54 elements in the periodic table.

It calculated force fields of thousands of nanosized clusters for each element in record time.

Said Argonne's Subramanian Sankaranarayanan, "The quality of our calculations for the 54 elements with the algorithm is much higher than the state of the art."

From Argonne National Laboratory
View Full Article


Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account