Brown CS News

A Team Led By Brown CS Master's Alum Prabhat Breaks The Exaflop Barrier And Wins ACM's Gordon Bell Prize


Click the link that follows to read more news about our pioneering and innovative alums.

"The CS community has talked about Exascale computing for a decade," says Brown CS Master's alum Prabhat, who recently received the ACM Gordon Bell Prize with colleagues from Lawrence Berkeley National Laboratory and NVIDIA, "and I'm thrilled that we were among the first ones to get there with a project that addresses one of the most important problems for all of us and our children."

Strongly contested on an international scale, the Gordon Bell Prize is awarded each year to recognize outstanding achievement in high-performance computing. It represents the highest honor for supercomputing experts, second perhaps only to the Turing Award. Prabhat and his team earned the prize by training a deep neural network to identify extreme weather patterns from high-resolution climate simulations, showing that accurate segmentation masks can be extracted for patterns such as tropical cyclones and atmospheric rivers.

"In the popular media, climate change is typically characterized by highly simplified quantities such as global, annual mean temperature or global sea level rise. Increasingly, the general public is interested in the impact of climate change and how climate change will affect them where they live,” Prabhat explains. “In order to address such questions, climate scientists need to develop capabilities for precision analytics: can we examine a 100-year climate simulation and automatically extract pixel-level segmentation masks corresponding to Hurricanes, Nor’easters and Atmospheric Rivers? It turns out that AI (and in particular Deep Learning) are up to the task, but there is a major performance bottleneck to overcome first. Our world-class team, comprising of researchers from NERSC, NVIDIA, and OLCF was able to adapt state-of-the-art segmentation architectures to work on million-pixel, 16-channel climate datasets, obtaining ~40TF performance on Volta GPUs, and 1.13 EF (16-bit) performance on the OLCF Summit system."

Prabhat began his career at Brown by earning his Master's degree in 2001, then spent more than a half-decade working at the Center for Computation and Visualization. "IMHO, there's no better school than Brown for breadth, depth, and exposure to other domain sciences," he says. "I was attracted to Brown by pioneers and strong mentors like Andy van Dam, who was always enthusiastic about learning himself and teaching others. David Laidlaw opened my eyes to the amazing, diverse possibilities of scientific visualization. I had the unique privilege of working with Thomas Banchoff and David on visualizing 4D mathematical objects in the CAVE for my Masters thesis. Upon graduation, Sam Fulcomer presented me with the  opportunity to work on diverse scientific projects at the CAVE, and CCV staff gave me room to grow and explore my interests. Working with Jim Head and Andy Forsberg on interactive visualization of large scale Martian topography and imagery data was probably the highlight of my time at CCV; I’d probably credit Jim the most for inspiring me to pursue a PhD in Earth Sciences at UC Berkeley."

Upon transitioning to Lawrence Berkeley National Lab, Prabhat's research interests shifted towards scientific visualization of large data sets, data management and supercomputing. He is currently the Data and Analytics Services Group Lead for the National Energy Research Scientific Computing Center (NERSC), where his group is responsible for deploying the Big Data stack for over 7000 science users.

“Our team is very proud of breaking the exaflop barrier and winning the Gordon Bell Prize. We’ve charted a new course for extreme scaling of Deep Learning architectures on HPC platforms. More than just a performance number, a research paper and an award, I’m personally most excited about the long-term possibilities for climate science. We now have a precision analytics tool that can help in assessing the impact of climate change for society at large. That is what really counts." Prabhat concludes.

You can also read the official ACM announcement, Lawrence Berkeley National Laboratory press release, and an NVIDIA blog post.

For more information, click the link that follows to contact Brown CS Communication Outreach Specialist Jesse C. Polhemus.