Bringing high-performance computing and machine learning closer to the action

It will likely be the first of its kind. In the world.

"No one else, to our knowledge, has tried to put such a high-speed computing cluster so close to the equipment in their labs," says Yaling Liu, a professor of bioengineering and mechanical engineering and mechanics in Lehigh University's P.C. Rossin College of Engineering and Applied Science. "We believe the facility we're proposing will have a huge impact on our campus by enabling big-data analysis and new science."

Liu leads the Lehigh team that recently received a nearly $1 million grant to fund the development of a heterogeneous edge computing platform for real-time scientific machine learning. The award is part of the National Science Foundation's Major Research Instrumentation program (MRI), which supports the development or acquisition of equipment that will advance science and engineering. One of the goals of such grants is to significantly impact research in a broad range of areas.

Joshua Agar, an assistant professor of mechanical engineering and mechanics at Drexel University, is also a member of the team, along with additional Lehigh contributors Lifang He, an assistant professor of computer science and engineering; Wujie Wen, an assistant professor of electrical and computer engineering; and Yue Yu, an associate professor of applied mathematics in Lehigh's College of Arts and Sciences.

Heterogeneous computing is a technique in which different processors are applied toward the execution of specific tasks, resulting in gains in performance or efficiency or both, while edge computing refers to the ability to process data as close to the data source as possible. Combining the two will address several critical computing needs.

"Computer clusters are typically located far from the equipment that generates the data," says Liu. "And that distance affects the speed by which data can be analyzed. So what ends up happening is that you collect all this data, and then you have to spend days analyzing it, after the fact. And if something went wrong with the experiment, you have to go back and repeat it, which can take hours or days, and can be expensive if you're paying hourly to use these machines."

Data storage poses another problem. Equipment like high-speed microscopes generate a huge amount of information, which means researchers like Liu are forced to keep their experiments short.

And finally, because of the lag between collection and analysis, the current system doesn't allow researchers to purge redundant or wasteful data. So valuable storage is sometimes taken up by information that's of no interest to them.

The proposed development of a heterogenous edge computing platform for real-time scientific machine learning will circumvent latency and bandwidth challenges, and allow for real-time analysis and control of optical, scanning probe, and transmission electron microscopy, thanks to its proposed location. Liu says the server will eventually be installed in the basement of the Health, Science, and Technology Building, with high-speed fibers connecting it to various basement labs housing the microscopes.

The platform will allow researchers across the university to make advances in fields like wireless communication, health care monitoring, bioinformatics, advanced manufacturing, and multiagent autonomous systems, and will facilitate interdisciplinary teams pursuing novel research directions.

For Liu, it could ultimately mean getting vital information to patients, faster.

Liu leads Lehigh's Bio-Nano Interface Lab, where he and his team currently use the high-speed camera of an optical microscope to sort tumor cells from normal cells. But with each blood sample containing hundreds of millions of cells, the researchers are limited by how fast they can sort the cells and how long their experiments can last. And they're unable to make decisions in the moment about what they're seeing.

"My lab is interested in the biomedical application of this platform," he says. "We want to make real-time decisions based on the image as to whether something is a tumor cell or a healthy white blood cell. That's what this facility is ultimately going to make possible. With this high-speed, edge computing, we'll be able to decide, 'This is a target cell,' and then we can collect it out immediately, and do further culturing or precision-based medicine based on those sorted cells."

What may be the world's first such computing platform configuration will most definitely be a game changer for him and his research.

"We'll be able to do continuous, real-time analysis that will enable us to finish our tasks in one day, rather than over several months," he says. "That will allow us to advance our research, which is incredibly exciting."

Provided by Lehigh University