PhD Proposal: Martin Schatz, GDC 4.416

Contact Name: 
Lydia Griffith
Date: 
Dec 6, 2013 3:00pm - 5:00pm

Ph.D. Oral Proposal: Martin Schatz

Date: December 6th, 2013
Time: 3:00 PM CST
Location: GDC 4.416

Supervising professor: Robert van de Geijn

Title: Anatomy of Computation with Tensors in Parallel

Abstract:

Recently, data models have become more complex leading to the need for multi-dimensional repre- sentations to express the data in a more meaningful way. Commonly, tensors are used to represent such data. Multi-linear algebra, the math associated with tensors, has become essential for tackling problems in big data and scientific computing. To solve the largest problems of today, libraries designed for super- computers consisting of thousands of compute nodes connected via a network are utilized. Such compute architectures are referred to as “distributed-memory” architectures. Up to now, the main approach for problems of multi-linear algebra has been based on mapping the multi-linear algebra to linear algebra and rely on highly efficient linear algebra libraries to perform the equivalent computation [5, 21]. Un- fortunately, there are inherent inefficiencies associated with this approach. In this proposal, we define a domain-specific language for distributed tensor computation. Additionally, through a process akin to constraint propagation, we show how, using the language, algorithms can be systematically derived and required collective communications identified for the tensor contraction operation.