Rob
BISSELING

PROJECT

"Data partitioning methods for parallel computations on large sparse matrices and tensors"

Parallel computers can speed up applications by using many processors simultaneously. For this, two challenging problems must be solved: balancing the computational work load across the processors and reducing the communication between them.

Mathematical objects called hypergraphs can be used to model computation and communication in irregular applications. A hypergraph represents a network of nodes (e.g. persons in a social network), together with groups of nodes (groups of friends).

A hypergraph can be stored as a sparse matrix, with many zero elements. Sparse matrices occur for instance in the study of biological, chemical, and social networks. Recent progress in artificial neural networks gave rise to higher dimensional analogues called sparse tensors.

The goal of this project is to obtain better data partitioning methods by combining the joint expertise in parallel sparse matrix and tensor computations of Rob Bisseling at Utrecht University and Bora Uçar at LIP in ENS  de Lyon.

Rob Bisseling is a full professor in Scientific Computing at the Mathematical Institute of Utrecht University, the Netherlands. His main research interest are: parallel algorithms, combinatorial scientific computing, graph and hypergraph algorithms, fast Fourier transforms. He is author of the book Parallel Scientific Computation: A Structured Approach using BSP, Second Edition, Oxford University Press 2020. The book is supported by a series of 26 educational videos available on his YouTube channel and by the educational parallel software package BSPedupack.

Rob Bisseling is the author or co-author of numerous open-source software packages: the Mondriaan sparse matrix partitioner; communication libraries BSPlib, MulticoreBSP, and Bulk; SAWdoubler for counting self-avoiding walks. His current main project is the development of PMondriaan, a parallel version of the Mondriaan sparse matrix partitioner. Another recent project is FFTU, a package for multidimensional parallel fast Fourier transforms (published Dec. 2023 in the SIAM Journal on Scientific Computing).

Rob Bisseling obtained BSc and MSc degrees in Mathematics from Radboud University, Nijmegen, in 1977 and 1981, and a PhD degree in Theoretical Chemistry from the Hebrew University, Jerusalem, in 1987. He was a research mathematician at the Shell laboratory in Amsterdam from 1987-1993, and since 1993 he works at Utrecht University.