Prof. John Gilbert (Computer Science Department, UCSB)

Title: Challenges in High-Performance Combinatorial Scientific Computing

Abstract: The world’s largest computers have always been used for simulation and data analysis to advance engineering capability and scientific knowledge. In the twentieth century, this usually meant numerical floating-point computation of continuous physical models based on differential equations. In this century, the largest computations deal with combinatorial structures: web links, DNA sequences, network topologies, relationship analysis, and so on. Combinatorial relationships among discrete entities are naturally modeled by graphs. However, the field of high-performance graph computation, in contrast with the mature field of numerical supercomputing, is in its infancy.

We are developing algorithms and toolboxes for graph analysis, with the goal of enabling scalable high-performance combinatorial computing. I will describe a few of our applications, outline our work on algebraic graph primitives, and talk about some of the challenges for the future of combinatorial computing.