May 10, Clark Fitzgerald (Student Seminar Series)

Student Seminar Series

DATE:              Wednesday May 10th, 2017, 11:10am

LOCATION:      MSB 1147 (Colloquium Room).

SPEAKER:         Clark Fitzgerald, (PhD Student, Statistics, UC Davis)

TITLE:              “Parallel Computing Through Code Analysis”

ABSTRACT:      Conventional systems for parallel programming require users to modify existing code to take advantage of a platform’s computational capabilities, which may include multiprocessing or Graphical Processing Units (GPU)s. In this presentation we review the state of the art for parallel programming in R and consider code analysis methods to detect the potential for parallel execution. The results of the analysis can then be used to rewrite and execute semantically equivalent parallel instructions, without requiring the user to modify their original code. We consider a motivating case study using robust regression to analyze the operating characteristics California’s highway traffic sensor stations on hundreds of gigabytes of traffic sensor data.