Machine learning program for games inspires development of breakthrough science tool

Newswise — A new AI tool models the behavior of clusters of nanoparticles in record time.

We learn new skills through repetition and reinforcement learning. Through trial and error, we repeat actions that lead to good results, try to avoid bad results, and seek to improve those in between. Researchers are now designing algorithms based on a form of artificial intelligence that uses reinforcement learning. They apply them to automate chemical synthesis, drug discovery, and even play games like chess and Go.

Scientists at the US Department of Energy’s (DOE) Argonne National Laboratory have developed a reinforcement learning algorithm for another application. It is used to model the properties of materials at the atomic and molecular scale and is expected to dramatically accelerate materials discovery.

Like humans, this algorithm “learns” problem solving from its mistakes and successes. But it does so without human intervention.

Historically, Argonne has been a world leader in molecular modeling. This involved calculating the forces between the atoms of a material and using that data to simulate its behavior under different conditions over time.

However, these earlier models relied heavily on human intuition and expertise and often required years of painstaking effort. Team reinforcement learning algorithm reduces time to days and hours. It also produces higher quality data than is possible with conventional methods.

“Our inspiration was AlphaGo,” said Sukriti Manna, research assistant at the Center for Nanoscale Materials (CNM) at Argonne, a DOE Office of Science user facility. “This is the first computer program to defeat a world champion Go player.”

The standard Go board has 361 positional squares, much larger than the 64 on a chessboard. This results in a large number of possible card configurations. The key to AlphaGo becoming a world champion was his ability to improve his skills through reinforcement learning.

Automating molecular modeling is of course very different from a Go computer program. “One of the challenges we faced is similar to developing the algorithm required for self-driving cars,” said Subramanian Sankaranarayanan, group leader at CNM Argonne and associate professor at the University of Illinois at Chicago.

While the Go map is static, traffic environments are constantly changing. The autonomous car must interact with other cars, varying routes, traffic signs, pedestrians, intersections, etc. Decision-making parameters are constantly changing over time.

Solving difficult real-world problems in materials discovery and design also involves continuous decision-making in the search for optimal solutions. The team’s algorithm includes decision trees that distribute positive reinforcement based on the degree of success in optimizing model parameters. The result is a model capable of accurately calculating material properties and their changes over time.

The team successfully tested their algorithm with 54 elements from the periodic table. Their algorithm learned to calculate the force fields of thousands of nanometer clusters for each element and performed the calculations in record time. These nanoclusters are known for their complex chemistry and the difficulty of traditional methods to model them accurately.

“It is like finishing the calculations of several doctorates. dissertations in days each, instead of years,” said Rohit Batra, a CNM expert on machine learning and data-driven tools. The team performed these calculations not only for single-element nanoclusters, but also for two-element alloys.

“Our work represents a major breakthrough in this type of model development for materials science,” Sankaranarayanan said. “The quality of our calculations for the 54 elements with the algorithm is far superior to the state of the art.”

Running the team’s algorithm required computations with large data sets on high-performance computers. To this end, the team used the CNM’s Carbon Cluster of Computers and the Theta supercomputer at the Argonne Leadership Computing Facility, a DOE Office of Science user facility. They also relied on the computing resources of the National Energy Research Scientific Computing Center, a user facility of the DOE Office of Science at Lawrence Berkeley National Laboratory.

“The algorithm should dramatically speed up the time needed to solve big challenges in many areas of materials science,” said CNM computational and theoretical chemist Troy Loeffler. Examples include materials for electronic devices, catalysts for industrial processes and battery components.

The team reported their findings in Nature Communications. Besides Sankaranarayanan, Manna, Batra, and Loeffler, Argonne contributing authors include Suvo Banik, Henry Chan, Bilvin Varughese, Kiran Sasikumar, Michael Sternberg, Tom Peterka, Mathew Cherukara, and Stephen Gray. Bobby Sumpter of Oak Ridge National Laboratory also contributed.

The work was supported by the DOE Office of Basic Energy Sciences.

The Argonne Leadership Computing Facility provides supercomputing capabilities to the scientific and engineering community to advance fundamental discovery and understanding across a wide range of disciplines. Supported by the Advanced Scientific Computing Research (ASCR) program of the U.S. Department of Energy’s (DOE) Office of Science, the ALCF is one of two DOE advanced computing facilities dedicated to open science.

Argonne National Laboratory seeks solutions to pressing national problems in science and technology. The nation’s first national laboratory, Argonne conducts cutting-edge basic and applied scientific research in virtually every scientific discipline. Argonne researchers work closely with researchers from hundreds of companies, universities, and federal, state, and municipal agencies to help them solve their specific problems, advance American scientific leadership, and prepare the nation for a better future. With employees in more than 60 countries, Argonne is managed by UChicago Argonne, LLC for the US Department of Energy’s Office of Science.

U.S. Department of Energy Office of Science is the largest supporter of basic physical science research in the United States and strives to address some of the most pressing challenges of our time. For more information, visit https://​ener​gy​.gov/​s​c​ience.

Sherry J. Basler