[Ml-stat-talks] Fwd: Reminder: Joint Seminar with Computer Science: Lester Mackey on Monday, December 14, 2015 at 12:30pm | Computer Science, Room 105
Barbara Engelhardt
bee at princeton.edu
Mon Nov 30 08:54:17 EST 2015
Talk of interest in two weeks.
---------- Forwarded message ----------
Subject: Reminder: Joint Seminar with Computer Science: Lester Mackey on
Monday, December 14, 2015 at 12:30pm | Computer Science, Room 105
To: CSML-seminars at princeton.edu
Lester Mackey-Stanford University
Joint Seminar with Computer Science
Monday, December 14, 2015
12:30pm-1:30pm
Computer Science, Room 105
**Lunch will be provided**
Title: “Matrix Completion and Matrix Concentration”
Abstract: The goal in matrix completion is to recover a matrix from a small
subset of noisy entries. Web-scale instances of this problem include
collaborative filtering for recommendation and link prediction in social
networks. Many modern matrix completion algorithms provide strong recovery
guarantees but scale poorly due to the repeated and costly computation of
truncated SVDs. To address this deficiency, in the first part of this talk,
I will introduce Divide-Factor-Combine (DFC), a parallel divide-and-conquer
framework for boosting the scalability of a matrix completion algorithm
while retaining its theoretical guarantees. Our experiments demonstrate the
near-linear to super-linear speed-ups attainable with this
approach, and our analysis shows that DFC enjoys high-probability recovery
guarantees comparable to those of its base algorithm.
Fundamental to our analysis – and to the analyses of
many matrix completion procedures – are matrix concentration inequalities
that characterize the fluctuations of a random matrix about its mean. In
the second part of this talk, I will show how Stein’s method of
exchangeable pairs can be used to derive concentration inequalities
for matrix-valued random elements. When applied to a sum of independent
random matrices, this approach yields matrix generalizations of the
classical inequalities due to Hoeffding, Bernstein,
Khintchine, and Rosenthal. The same technique delivers bounds for sums of
dependent random matrices and more general matrix functionals of dependent
random elements.
If you would like to be added to the CSML seminar listerv, please email
capizzi at princeton.edu
Joseph D. Capizzi Jr.
Administrative Assistant to the Director
Center for Statistics and Machine Learning
Green Hall, 3-C-5
Princeton University
Princeton, NJ 08544
capizzi at princeton.edu
609-258-9862
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20151130/8961f420/attachment.html>
More information about the Ml-stat-talks
mailing list