CMU and PSC Researchers Win HPCwire Awards

Byron SpiceFriday, December 5, 2014

Computer Science Professor Tuomas Sandholm, Ph.D. student John Dickerson and the Pittsburgh Supercomputing Center won the HPCwire Reader's Choice Award for Best Data-Intensive Application/Use for their work to optimize living-donor kidney exchanges.

Tuomas Sandholm, professor of computer science, and his Ph.D. student, John Dickerson, along with the Pittsburgh Supercomputing Center, won the HPCwire Reader's Choice Award for Best Data-Intensive Application/Use for their work with PSC's Blacklight supercomputer to optimize living-donor kidney exchanges.

They investigated how to optimize current kidney exchanges using distributional knowledge of the future state of the kidney exchange pool, and how to incorporate human expert value judgment into a concrete, numeric optimization objective using data.

Their algorithm matches living kidney donors with medically compatible transplant candidates, enabling more paired-donor transplants. In such transplants, each donor-recipient pair includes a potential donor who is not medically compatible with his or her original intended recipient, but is compatible with another recipient who likewise has an incompatible donor.

The group tied with IBM's question-answering Watson for the prize, which was presented at the 2014 International Conference for High Performance Computing, Networking, Storage and Analysis (SC14) in New Orleans.

Sandholm's team successfully used Blacklight, PSC's large shared-memory supercomputer supported by the National Science Foundation's Division of Advanced Cyberinfrastructure, to test FutureMatch, a data-driven framework for learning. FutureMatch relies on extensive computational simulations of long-running dynamic kidney exchanges to learn its matching strategy.

Simulating kidney exchange is data-intensive, requiring tens of thousands to millions of runs to ensure statistical significance. Blacklight's huge shared coherent memory — up to 16 terabytes — allows FutureMatch to solve large kidney exchange problems in parallel without exhausting memory.

PSC and UCLA also won a second Reader's Choice Award for Best Use of HPC Application in Life Sciences for using the PSC's ANTON supercomputer to develop cholesterol-lowering drugs.

"HPCwire readers are among the most informed in the HPC community and these awards are ultimately given to the organizations that are making the biggest impacts in advancing technology and humanity itself through high-performance computing," said Tom Tabor, CEO of Tabor Communications, publisher of HPCwire. More information on the awards is available on the HPCwire website.

For More Information

Byron Spice | 412-268-9068 | bspice@cs.cmu.edu