Building accurate models of climate, mechanics, and space requires advanced mathematics. That's where people like Idaho State University professor Yuri Gryazin and his students step in.
Recently, Gryazin, an expert in numerical analysis and scientific computing, and his collaborators at Berkeley Lab in California and Rice University in Texas began researching new machine learning algorithms related to simulation and computer modeling. . To support this effort, the U.S. Department of Energy awarded the team his $4 million grant under the newly established “Scientific Machine Learning for Complex Systems” program. Digital models are used throughout his STEM fields to simulate everything from the movement of Earth's weather and seismic waves to the operation of particle accelerators and nuclear reactors.
“Practically all new developments are first modeled using advanced numerical approaches,” Gryazin said. “Recently, major advances in machine learning algorithms have opened new directions for developing faster and more accurate computational methods for such simulations.”
Gryazin and his students focus on the uncertainty quantification part of the simulation problem. Common uncertainty quantification issues of interest include authentication, prediction, model and software validation and validation, parameter estimation, and data assimilation. Quantifying uncertainty helps researchers determine how reliable a model's predictions are.
“You've probably seen surveys and polls that have a margin of error of plus or minus a certain percentage,” Gryazin said. “Uncertainty quantification is similar, except that it involves models and simulations.”
Because the development of many models requires data from potentially millions or more data points, finding the margin of error for these scientific models is a leap beyond the average survey. becomes more complicated. Using advanced techniques in applied mathematics and statistics, Gryazin and his students create and train new neural network algorithms (a type of machine learning structured like the human brain) based on known data. and then test that approach. During testing, an algorithm's success or failure depends on how accurately it predicts the correct outcome when given a new real-world data set that was not present in the initial training set. Masu. Specifically, it works with algorithms used in underground imaging systems, such as those used to detect mines, air pockets, contaminants, etc. Three of her students from multiple disciplines at Idaho State University will work on this project.
“This new opportunity to collaborate with experts from a respected institution like Berkeley Lab provides a valuable learning opportunity,” said Adil Ahmed, a third-year mechanical engineering major. “Also, collaborating with leading researchers in this field makes this project particularly exciting to be involved in.”
The group hopes to publish first results by the end of 2024, when the first new underground neural network algorithms will be developed and tested.
“Students working on this project will gain valuable experience collaborating with world-class researchers from the nation's top scientific centers to solve important problems in exciting new areas of research,” Gryazin said. Stated.
For more information about the Department of Mathematics and Statistics, please visit the following website: isu.edu/mathematics.
