CAM Colloquium: Alan Edelman (MIT) - The GSVD, BackProp, & Julia

Location

Frank H. T. Rhodes Hall 655

Description

Abstract:
In this “something for everybody” talk, I will begin by discussing the Generalized Singular Value Decomposition, invented right here at Cornell in 1976 by Charlie van Loan. Many people have wondered whether there is an ellipse picture for the GSVD. We show that in 4 dimensions or more, one can truly appreciate the GSVD picture. We will then discuss how you might find an application of the GSVD of your own. In the second part of my talk, I will explain Back Propagation, not through the chain rule but through basic numerical linear algebra algorithms. What makes this explanation interesting is that it can be readily implemented in Julia using Julia’s generic algorithm approaches avoiding the need to rewrite the basic algorithms. This is joint work with Bernie Wang, and Ekin Akyurek.

Bio:
Alan Edelman is Professor of Applied Mathematics, and in 2004 founded Interactive Supercomputing (acquired by Microsoft). He received the B.S. & M.S. degrees in mathematics from Yale in 1984, and the Ph.D. in applied mathematics from MIT in 1989 under the direction of Lloyd N. Trefethen. Following a year at Thinking Machines Corp and at CERFACS in France, Edelman went to U.C. Berkeley as a Morrey Assistant Professor and Lewy Fellow, 1990-93. He joined the MIT faculty in applied mathematics in 1993. Edelman's research interests include high performance computing, numerical computation, linear algebra and stochastic eigenanalysis (random matrix theory). He has consulted for Akamai, IBM, Pixar, and NKK Japan among other corporations. A Sloan fellow, Edelman received an NSF Faculty Career award in 1995. He has received numerous awards, among them the Gordon Bell Prize and Householder Prize (1990), the Chauvenet Prize (1998), the Edgerly Science Partnership Award (1999), the SIAM Activity Group on Linear Algebra Prize (2000), and the Lester R. Ford Award, (2005). In 2011, Edelman was selected a Fellow of SIAM, "for his contributions in bringing together mathematics and industry in the areas of numerical linear algebra, random matrix theory, and parallel computing." Edelman was named a 2018 Fellow of the IEEE for his "contributions to the development of technical-computing languages," namely the Julia language for numerical/scientific computing.