Abstract: Tensor decomposition is a common approach for analysis of multiway datasets, e.g., object-by-feature-by-scenario. Tensor decomposition is an unsupervised learning technique that can be used for latent factor analysis, dimensionality reduction, compression, missing data estimation, and more. In this talk, we focus on the CANDECOMP/PARAFAC (CP) tensor decomposition, also known as the canonical polyadic decomposition. We discuss methods for fitting the CP model to real-world datasets and how randomization can be used to accelerate the process. Specifically, we look at matrix sketching techniques for least squares problems and stochastic gradient descent. We also consider a generalization of the CP model that can incorporate alternative objective functions, such as analogues of logistic loss. We show real-world examples to demonstrate the approach. This is joint work with Grey Ballard (Wake Forest), Casey Battaglino (Georgia Tech), Jed Duersch (Sandia), David Hong (Michigan), and Brett Larsen (Stanford), and Alex Williams (Stanford).
Bio: Dr. Kolda is a Distinguished Member of the Technical Staff at Sandia National Laboratories. She has led numerous projects in computational science and data analysis on topics in multilinear algebra and tensor decompositions, graph models and algorithms, data mining, optimization, nonlinear solvers, parallel computing and the design of scientific software. Her work has received several honors, including a 2003 Presidential Early Career Award for Scientists and Engineers (PECASE), an R&D100 award, and three best paper prizes at international conferences. She was named a Distinguished Scientist of the Association for Computing Machinery (ACM) in 2011 and a Fellow of the Society for Industrial and Applied Mathematics (SIAM) in 2015. She is currently a member of the SIAM Board of Trustees and is the founding Editor-in-Chief of the new SIAM Journal on Mathematics of Data Science.
The Robert Mueller-Thuns Lectureship in Computer Science was established at the University of Illinois through memorial gifts by his wife, Nikki Mueller-Thuns Mirghafori (BS '91), family, friends, and colleagues.
Robert was born in Bonn, Germany and did his undergraduate work at the Technical University of Aachen, Germany. He attended University of Illinois at Urbana-Champaign from 1986 to 1990, where he received both his M.S. and Ph.D. degrees in Computer Science. While in graduate school, he held internships at IBM T.J. Watson Research Center in Yorktown Heights, followed by positions at Cadence Design Systems and TCSI Corporation post-graduation. At the time of his untimely death at the age of 37, he was a Software Architect at Evolve Corporation, a start-up in San Francisco, CA. His areas of expertise were computer-aided design for VLSI, simulation, testing, parallel processing, formal design verification, as well as object oriented analysis and design, and pattern languages.
Robert was a remarkable human being: a gentle spirit with an insatiable intellectual curiosity and zest for life. Aside from being a researcher and computer scientist, he was a musician, an athlete, and a writer. He had a passion for thinking, ideas, and learning, as well as for living life through friendship, music, food, and travel. Even as he battled cancer, he never lost his optimism, sense of humor, and vitality. In the words of Paul Chen (PhD EE '93), a close friend: "Robert was known by his Illini friends as The Uberman, partly because of his German ethnicity, and partly because he was an amazing individual—someone whom we respected and admired for his ability to achieve excellence in all that he did. He had a passion for living. And he excelled in life: the Uberman could out-almost-anything us, all the while smiling effortlessly and encouragingly at us lesser mortals."
Robert thrived and made life-long friends at Illinois and it is hoped that the Mueller-Thuns Lectureship in Computer Science will give the chance to others at Illinois to think more, learn more, and do more as an appropriate memorial to a man whose intellectual intensity, integrity, and vibrancy made him an inspiration to all who knew him.