A0774
Title: High-performance computing for sparse graphical models
Authors: Joong-Ho Won - Seoul National University (Korea, South) [presenting]
Abstract: High-performance computing (HPC) means computations that are so large that a single (desktop) computer cannot meet their requirement on storage, main memory, and raw computational speed. Computational issues involved with graphical model selection in ultrahigh-dimensional data are discussed, in which the number of variables is about a million. At this scale, HPC on distributed memory systems is often necessary. Communication-avoiding matrix multiplication plays a central role in this setting, provided that a proper method is employed. The experience with communication-avoiding algorithms in two national supercomputing centres is shared.
