Aditya Thakur receives DOE Early Career Award
Computer Science (CS) Assistant Professor Aditya Thakur has received an Early Career Award from the U.S. Department of Energy's Office of Science. The annual awards support the research programs of outstanding scientists and engineers who are early in their careers with a $150,000 each year for summer salary and research expenses.
Thakur is one of three UC Davis researchers to receive the award this year and one of 83 nationwide. He joins his colleague, CS associate professor Cindy Rubio González, electrical and computer engineering professor John Owens and materials science and engineering faculty Ricardo Castro and Roopali Kukreja as the only College of Engineering faculty to receive the award.
Thakur's proposal, "AutoNeurify: Automatic Infusion of Learning in HPC Applications" will help improve the performance of high-performance computing (HPC) applications by using machine learning while lowering the level of expertise required for their maintenance and deployment.
HPC applications are essential for scientific progress, as computational methods for data collection and simulation are critical in making foundational discoveries in nearly all scientific disciplines. HPC applications are also essential for addressing humanity’s big problems, such as understanding and mitigating climate change, helping solve the energy crisis, and aiding in drug discovery. Consequently, improving the speed of scientific software and making it easier to deploy and maintain can help accelerate scientific discovery.
Recent research has shown that machine learning can speed up HPC applications, but it puts a significant burden on scientists. In particular, integrating machine learning into existing HPC applications requires significant machine-learning expertise and manual effort. Exploiting the latest advances in machine learning requires a tightly-coupled integration of machine learning into existing HPC workflows.
To address this need, Thakur plans to develop AutoNeurify, an end-to-end system to automatically infuse learning in HPC applications. The program will enable scientists to access a powerful machine-learning toolkit while reducing software-maintenance costs and improving maintainability and portability of software. The project draws upon Thakur's expertise in programming languages, formal methods and machine learning, all of which are critical for this project.
"This is an interdisciplinary project that will use advances in multiple areas of computer science to build a tool that will help scientists more effectively use the latest advances in machine learning," Thakur said. “We are especially excited about using AutoNeurify to speed up geophysical fluid dynamics simulations for climate change modeling.”
Since joining UC Davis in 2017, Thakur has received an NSF CAREER Award 2021, the Facebook Probability and Programming Research Award 2019 and 2020 and the Facebook Testing and Verification Research Award 2018.