Presenter
Abstract
Algorithms have two costs: arithmetic and communication, i.e. moving data between levels of a memory hierarchy or processors over a network. Communication costs (measured in time or energy per operation) already greatly exceed arithmetic costs, and the gap is growing over time following technological trends. Thus the goal is to design algorithms that minimize communication. The speaker will present new algorithms that communicate asymptotically less than their classical counterparts, for a variety of linear algebra and machine learning problems, demonstrating large speedup on a variety of architectures. Some of these algorithms attain provable lower bounds on communication. The speaker will also describe a generalization of these lower bounds and optimal algorithms to arbitrary code that can be expressed as nested loops accessing arrays, assuming only that array subscripts are affine functions of the loop indices, a special case being convolutional neural nets.
Additional Information
About the Speaker: James Demmel is the Dr. Richard Carl Dehmel Distinguished Professor of Computer Science and Mathematics at the University of California at Berkeley, and chair of the EECS Department. His research is in numerical linear algebra, high-performance computing, andcommunication-avoiding algorithms.
About the Lecture Series: Begun in 2014, theHouseholderLectureSeries honors Dr. Alton S.Householder, who spent most of his career in the ORNL Mathematics Division (now the Computational Science and Mathematics Division), serving as the director for much of that time, before joining the Department of Mathematics at the University of Tennessee. The series is sponsored jointly by the ORNL Computational and Applied Mathematics Group and the Department of Mathematics at the University of Tennessee, Knoxville. Its purpose is to have distinguished scientists present their research on various topics in computational and applied mathematics.