UW CSE Ph.D. student Maaz Bin Safeer Ahmad, who works with professor Alvin Cheung of UW CSE’s Programming Languages & Software Engineering (PLSE) and Database groups, has captured the inaugural Best Student Paper Award at SYNT 2016 for Leveraging Parallel Data Processing Frameworks with Verified Lifting.
In the winning paper, Ahmad and Cheung demonstrate that verified lifting—which previously has been applied to database applications and stencil computations—also can be used to convert sequential data processing code in order to leverage high-performance parallel data processing frameworks. They present CASPER, a novel compiler that identifies and converts fragments of sequential Java code to MapReduce tasks implemented with Apache Hadoop.
The pair’s work represents a significant step forward in addressing one of the most pressing computational challenges of our time: the efficient processing and analysis of increasingly massive data sets. By enabling automatic rewriting of code written in general purpose languages into a high-performance framework such as Hadoop, the authors have devised a way for users to leverage the performance improvements offered by domain specific languages without having to expend time and resources on manually rewriting programs—and avoiding the risk of introducing new bugs. Evaluating their prototype using a set of standard MapReduce benchmarks, Ahmad and Cheung demonstrated that programs that have been optimized using CASPER run up to 6.4x faster.
Ph.D. students Sarah Chasins of UC Berkeley and Julie Newcomb of UW CSE earned second place for “Using SyGuS to Synthesize Reactive Motion Plans.” In their paper, Chasins and Newcomb—who work with professor Ras Bodik of UW CSE’s PLSE group—present the first use of the Syntax-Guided Synthesis (SyGuS) formalism to solve robot motion planning problems.
The SYNT Workshop on Synthesis is co-located with the International Conference on Computer Aided Verification (CAV 2016) taking place July 17-23 in Toronto, Canada.
Way to go, team!