[Ml-stat-talks] Wed: Jean Honorio on stochastic optimization

David Mimno mimno at CS.Princeton.EDU
Mon Nov 5 10:16:59 EST 2012


Over the past few years we've become more and more interested in stochastic
optimization, both for theoretical reasons and also as a route to that
buzziest of buzzwords, Big Data. This week we have MIT postdoc Jean
Honorio-Carillo, who will discuss bounds on optimization in Ising models.

Jean Honorio-Carillo, MIT
CS402, Wed Nov 7, 12:30

Title: Convergence Rates of Biased Stochastic Optimization for Learning
Sparse Ising Models

Abstract:
We study the convergence rate of stochastic optimization of exact (NP-hard)
objectives, for which only biased estimates of the gradient are available. We
motivate this problem in the context of learning the structure and
parameters of Ising models. We first provide a convergence-rate analysis of
deterministic errors for forward-backward splitting (FBS). We then extend
our analysis to biased stochastic errors, by first characterizing a family
of samplers and providing a high probability bound that allows
understanding not only FBS, but also proximal gradient (PG) methods. We
derive some interesting conclusions: FBS requires only a logarithmically
increasing number of random samples in order to converge (although at a
very low rate);
the required number of random samples is the same for the deterministic and
the biased stochastic setting for FBS and basic PG;
accelerated PG is not guaranteed to converge in the biased stochastic
setting.

Bio:
Jean Honorio is a PhD in Computer Science from Stony Brook University. He
received his MSc in Computer Science from The George Washington University
in 2006, and BSc in Systems Engineering from Universidad de Lima, Peru in
1997. His research focuses on tractable learning of graphical model
structures from data. He will soon join CSAIL at MIT as a post-doctoral
researcher. He was member of the Image Analysis Lab at Stony Brook
University and the Medical Department at Brookhaven National Lab in
2007-2012, the Institute for Computer Graphics at The George Washington
University in 2005-2006, and the Institute for System Engineering and
Operations Research at Universidad de Lima in 1995-1996.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20121105/5dafa54c/attachment.html>


More information about the Ml-stat-talks mailing list