You are here

ECE MS Thesis Defense: "Distributing Frank-Wolfe via Map-Reduce," Armin Moharrer


128 FR

April 18, 2018 9:30 am
April 18, 2018 9:30 am

Large-scale optimization problems abound in data mining and machine learning applications, and the computational challenges they pose are often addressed through parallelization. We identify structural properties under which a convex optimization problem can be massively parallelized  via map-reduce operations using  the Frank-Wolfe (FW) algorithm. The class of problems that can be tackled this way is quite broad and includes experimental design, AdaBoost, and projection to a convex hull.  

Implementing FW via map-reduce  eases parallelization and deployment via commercial distributed computing frameworks. We demonstrate this by implementing FW  over Spark, an engine for parallel data processing, and establish that parallelization through map-reduce yields significant performance improvements: we solve problems with 10 million variables using 350 cores in 44 minutes; the same operation takes 133 hours when executed serially.

  • Professor Stratis Ioannidis (Advisor)
  • Professor Jennifer Dy
  • Professor David Kaeli