Tanzima Islam’s innovative computing work garners Early Career Research Program award
Tanzima Islam, an assistant professor in the Department of Computer Science at Texas State University, has been named recipient of the U.S. Department of Energy’s Office of Science’s Early Career Research Program, which bolsters the nation’s scientific workforce by supporting exceptional researchers at the outset of their careers, when many scientists do their most formative work.
The five-year, $770,000 grant will support Islam’s research project, “INTELYTICS: An Efficient Data-Driven Decision-Making Engine for Performance in the Era of Heterogeneity,” which seeks to build generalizable and interpretable machine learning (ML) techniques for recommending parameter settings to users, software and facilities so that simulations can finish faster by efficiently utilizing the computing resources.
Islam is one of 83 early career scientists from across the country selected by the DOE for the Early Career Research Program, representing 47 universities and 13 national labs in 29 states. These awards are a part of the DOE’s long-standing efforts to develop the next generation of STEM leaders who will solidify America’s role as the driver of science and innovation around the world.
Since its inception in 2010, the Early Career Research Program has made 785 awards, with 508 awards to university researchers and 277 awards to National Lab researchers.
Islam’s project will address issues of efficiency in high performance computing (HPC). Simulations leverage HPC systems to quickly evaluate numerous what-if scenarios of otherwise experimentally intractable phenomena. However, HPC ecosystems consist of diverse components, and failure to consider how variables at the user, software, or system level interact when running these simulations can delay the scientific return or leave billion-dollar systems underutilized.
Although artificial intelligence(AI)/ML-based performance modeling and tuning can automate the decision-making process, off-the-shelf ML techniques struggle to do well for scientific data, especially when environments change and a small amount of training data is available. Consequently, new approaches are needed to address the gap unique to performance analytics in the context of making predictions quickly for volatile environments without requiring substantial training data.
Islam’s research would overcome these shortcomings by innovatively transforming data and leveraging the recent advancements in AI/ML techniques to accelerate real-time decision-making in HPC. The resulting engine, INTELYTICS, will address the shortcomings identified by Advanced Scientific Computing Research (ASCR) experts toward an automated workflow management system in heterogeneous computing environments.
For more information, visit the Early Career Research Program webpage.