[Ml-stat-talks] Princeton Optimization Seminar: Donald Goldfarb, Columbia University

Amir Ali Ahmadi a_a_a at princeton.edu
Fri Apr 10 11:12:13 EDT 2015

-----   Princeton Optimization Seminar   -----

DATE:  Thursday, April 16

TIME:  4:30 pm

LOCATION:  Sherrerd Hall 101

SPEAKER: Donald Goldfarb, Columbia University

TITLE:  Low-rank Matrix and Tensor Recovery: Theory and Algorithms<https://orfe.princeton.edu/abstracts/optimization-seminar/low-rank-matrix-and-tensor-recovery-theory-and-algorithms>

Recovering a low-rank matrix from incomplete or corrupted observations is a recurring problem in processing and machine learning. For problems in which the intrinsic structure of incomplete or corrupted data is more than 3-dimensional, low-rank completion and RPCA convex models for matrices have been extended to tensors. Here we establish recovery guarantees for both tensor completion and tensor RPCA, show that using the most popular convex relaxation for the tensor Tucker rank can be substantially sub-optimal in terms of the number of observations needed for exact recovery and introduce a very simple new convex relaxation that is theoretically and empirically much better. We also propose algorithms to solve these models that are based on Alternating Direction Augmented Lagrangian (ADAL), Frank-Wolfe and prox-gradient methods, and empirically study their performance on both simulated and real data.

*This is joint work with: Cun Mu, Bo Huang and Tony Qin {current and former IEOR PhD students at Columbia University} and John Wright {E.E. faculty member at Columbia University}.


Professor Goldfarb is internationally recognized for his contributions to the field of optimization, and in particular for the development and analysis of efficient and practical algorithms for solving various classes of optimization problems. His most well-known and widely used algorithms include steepest-edge simplex algorithms for linear programming, the BFGS quasi-Newton method for unconstrained optimization, and the Goldfarb-Idnani algorithm for convex quadratic programming. He has also developed simplex and combinatorial algorithms for network flow problems, and interior-point methods for linear, quadratic and second-order cone programming. His recent work on robust optimization for portfolio selection, algorithms for image de-noising, compressed sensing and machine learning is very highly cited.

Professor Goldfarb joined the IEOR Department at Columbia in 1982, serving as Chair for eighteen years from 1984-2002, after having spent 14 years on the faculty of the Department of Computer Sciences, which he co-founded, at the City College of New York. He served as Interim Dean of Columbia’s School of Engineering and Applied Science during the 1994-95 and 2012-13 academic years and its Executive Vice Dean in the Spring 2012 semester. He currently holds the Avanessians Chair in IEOR and is a SIAM Fellow. He was awarded the Khachiyan Prize in 2013 and the Prize for Research Excellence in the Interface between OR and CS in 1995 by INFORMS, and is listed in The World’s Most Influential Scientific Minds, 2014, as being among the 99 most cited mathematicians between 2002 and 2012.

If you would like to subscribe to the mailing list of the optimization seminar series, please visit the link below or send an email with
SUBSCRIBE opt-seminar in the body to listserv at lists.princeton.edu<mailto:listserv at lists.princeton.edu>.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20150410/0d2f71d3/attachment.html>

More information about the Ml-stat-talks mailing list