[Ml-stat-talks] FW: [ORFE seminars] Wilks Statistics Seminar: Nathan Srebro, Today, Mar. 13th at 12:30pm, Sherrerd Hall 101

Han Liu hanliu at Princeton.EDU
Fri Mar 13 09:10:55 EDT 2015

Dear Colleagues,

         This talk is ML/STATS related, and may be of interest to many of you.

Best regards,

Han Liu,
Assistant Professor
Department of Operations Research and Financial Engineering
Princeton University, Princeton, NJ 08544
From: ORFE Talks [ORFE-TALKS at Princeton.EDU] on behalf of Carol Smith [carols at PRINCETON.EDU]
Sent: Friday, March 13, 2015 9:09 AM
To: ORFE-TALKS at Princeton.EDU
Subject: [ORFE seminars] Wilks Statistics Seminar: Nathan Srebro, Today, Mar. 13th at 12:30pm, Sherrerd Hall 101

***   Wilks Statistics Seminar   ***

DATE:   Today, March 13, 2015

TIME:   12:30pm

LOCATION:   Sherrerd Hall, room 101

SPEAKER:   Nathan Srebro, Toyota Technological Institute at Chicago

TITLE:   A Tale of Two Norms

ABSTRACT:  There has been much interest in recent years in various ways of constraining the complexity of matrices based on factorizations into a product of two simpler matrices. Such measures of matrix complexity can then be used as regularizers for such tasks as matrix completion, collaborative filtering, multi-task learning and multi-class learning. In this talk I will discuss two forms of matrix regularization which constrain the norm of the factorization, namely the trace-norm (aka nuclear-norm) and the so-called max-norm (aka γ 2 :ℓ 1 →ℓ ∞ norm). I will both argue that they are independently motivated and often better model data then rank constraints, as well as explore their relationships to the rank. In particular, I will discuss how simple low-rank matrix completion guarantees can be obtained using these measures, and without various "incoherence" assumptions. I will present both theoretical and empirical arguments for why the max-norm might actually be a better regularizer, as well as a better convex surrogate for the rank.

Based on joint work with Rina Foygel, Jason Lee, Ben Recht, Russ Salakhutdinov, Ohad Shamir, Adi Shraibman and Joel Tropp and others.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20150313/5d5b533d/attachment.html>

More information about the Ml-stat-talks mailing list