Sandia National Laboratories scientist Rémi Dingreville co-developed a way to speed up virtually any type of computer simulation, a ubiquitous research tool in science and industry. Credit: Craig Fritz
A good machine learning algorithm is a powerful research accelerator. Combined with computer simulation, it can detect mathematical shortcuts in the program, allowing scientists to more quickly gain insights into the effects of drugs on cells or the potential of rocket engines to send humanity to Mars and beyond.
New research is making this tool available to scientists around the world. In a paper recently published in npj computer materialsA team of researchers from Sandia National Laboratories and Brown University has introduced a universal method to speed up virtually any type of simulation.
“From a user perspective, there’s no difference between running your simulation tool or running this accelerated simulation tool. It gives you the exact same predictions. The difference is the time it takes to get those results,” said Sandia’s Rémi Dingreville.
Dingreville and his team ran a materials science simulation 16 times faster than normal with their accelerator. And more importantly, they explained in their paper how it can just as easily speed up computer programs for climate change research, autonomous vehicle navigation, or hardware acceleration.
“The potential for generalizing our approach to different systems could lead to more efficient and sustainable technologies,” said Brown’s Vivek Oommen, first author of the paper.
Accelerator democratizes fast science
As a child, Dingreville loved to go fast. He biked, skied, and ran. He even competed in races to be the first to finish his homework. Today, as a scientist, he uses machine learning to speed up his research. In a previous project, he redesigned a simulation to run 40,000 times faster.
While a 16-fold speedup may seem modest in comparison, Dingreville and his team point out that their latest work could have a much greater impact, because it benefits virtually every field of science. It is not limited to specific types of problems like other accelerators.
“Physics, chemistry, geochemistry, weather forecasting: it really doesn’t matter,” Dingreville said.
The team sees its paper as a challenge to researchers to fundamentally rethink how they design and use simulations.
“I am deeply fascinated by the challenges and potential of integrating traditional numerical methods with artificial intelligence to solve complex problems in materials science,” Oommen said.
Faster simulations open new research opportunities
While the simulation accelerator saves time and money for routine research, it also removes barriers to studying phenomena that normally can’t be simulated. Try to model a slow-moving event, like melting glaciers, and your program will likely take too long to be useful.
“The current state of the art is such that you have to use direct numerical solvers. Even if they are accurate, they are slow,” Dingreville said.
The team hopes that this research will be the genesis of a modern, common method for scientists to run normally slow simulations.
“In the future, I look forward to seeing how our methodologies can be applied to other complex problems in diverse fields, such as energy, biotechnology and environmental sciences,” Oommen said.
“I would like to see this applied to geoscience,” Dingreville added.
More information:
Vivek Oommen et al., Rethinking Materials Simulations: Combining Direct Numerical Simulations with Neural Operators, npj computer materials (2024). DOI: 10.1038/s41524-024-01319-1
Provided by Sandia National Laboratories
Quote:Universal accelerator finds faster answers to complex problems (2024, August 28) retrieved August 28, 2024 from
This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.
