This shows you the differences between two versions of the page.

Link to this comparison view

seminars:seminar_3_22_17 [2017/09/20 22:02] (current)
Line 1: Line 1:
 +====== Selectively Consistent Approximate Parallel Execution ======
 +Friday March 24, 2017\\
 +Location: CIC Panther Hollow Room\\
 +Time: 4:30PM\\
 +**Vignesh Balaji (CMU)**\\
 +Programmers writing shared memory parallel applications use synchronization
 +operations for correct manipulation of shared data. They also rely on hardware
 +based cache coherence to automatically transport data between cores. While
 +synchronization and cache coherence are required for correct execution of
 +parallel applications,​ they impose a significant performance penalty. The
 +serialization and data movement overheads of these operations impact
 +scalability of parallel applications. Research in approximate computing has
 +demonstrated the error tolerance of many important application domains by
 +improving performance with minimal impact on output quality. In this talk, I
 +will present our work on SCAPE - a system that exploits this resilience
 +towards errors to eliminate synchronization and cache coherence for
 +performance. Our system allows the programmer to control the quality of such
 +approximate executions by ensuring precise updates only for program values
 +deemed quality critical by the programmer. Additionally,​ ¬†SCAPE uses a neural
 +network to selectively approximate executions only when the expected
 +performance improvement justifies approximation. Our evaluations show that
 +SCAPE can improve performance up to 20X while keeping the quality degradation
 +less than 1% for a collection of graph applications.
 +Vignesh Balaji is a second year PhD student co-advised by Professors Brandon
 +Lucia and Radu Marculescu. His research is focused on approximate computing
 +with a particular emphasis on reducing data movement in multi-core processors.
 +**[[seminars| Back to the seminar page]]**