BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:The tensor graphical lasso (Teralasso) - Alfred Hero (University o f Michigan) DTSTART:20171031T140000Z DTEND:20171031T145000Z UID:TALK94117@talks.cam.ac.uk CONTACT:INI IT DESCRIPTION:Co-authors: Kristjian Greenewald (Harvard University)\, Shuheng Zhou (University of Michigan)\, Alfred Hero (University of Mich igan)

We propose a new ultrasparse graphical model f or representing multiway data based on a Kronecker sum representation of t he process inverse covariance matrix. This statistical model decomposes th e inverse covariance into a linear Kronecker sum representation with spars e Kronecker factors.

Under the assumption that the multiway obse rvations are matrix-normal the l1 sparsity regularized log-likelihood fun ction is convex and admits significantly faster statistical rates of conve rgence than other sparse matrix normal algorithms such as graphical lasso or Kronecker graphical lasso.

We specify a scalable composite grad ient descent method for minimizing the objective function and analyze both the statistical and the computational convergence ratesm\, showing that t he composite gradient descent algorithm is guaranteed to converge at a geo metric rate to the global minimizer. We will illustrate the method on seve ral real multiway datasets\, showing that we can recover sparse graphical structures in high dimensional data.

Related Links

https://arxiv.org/abs/1705.03983 - Pa per

LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR