From macglashan at tti-c.org  Mon Aug 18 09:14:00 2008
From: macglashan at tti-c.org (Julia MacGlashan)
Date: Mon Aug 18 10:06:42 2008
Subject: [TTIC Colloquium] Tong Zhang, Rutgers University- TTI-C Talk
Message-ID: <000001c9013c$af1d5350$aabf8780@jmacglDPLFYD1>

When:              Monday, August 25 @ 2:00pm

 

Where:            TTI-C Conference Room: 1427 E. 60th St, 2nd Floor

 

Who:                Tong Zhang: Statistics Department, Rutgers University

 

Topic:              Approximate Solution of L0 Regularization

 

 

The problem of minimizing a convex loss function under L0 regularization is
non-smooth and non-convex. In the general case, this problem is NP hard.
However, a number of recent work showed that the problem can be
approximately solved using convex relaxation under certain assumptions. In
this talk we present a more direct approach to this problem. I will
introduce a notion of approximate local minimum for L0 regularization. It is
shown that adaptive forward-backward greedy algorithms can efficiently
compute such a local minimum.

Moreover, under appropriate assumptions, the local minimum computed by
adaptive greedy procedures approximately solves the global optimization
problem.

 

Contact:          Shai Shalev-Shwartz, TTI-C         shai@tti-c.org
834-6850

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20080818/18bff1ba/attachment.htm
From macglashan at tti-c.org  Fri Aug 22 08:35:35 2008
From: macglashan at tti-c.org (Julia MacGlashan)
Date: Fri Aug 22 09:27:38 2008
Subject: [TTIC Colloquium] Tong Zhang, Rutgers University- TTI-C Talk
Message-ID: <000001c9045b$f583f250$aabf8780@jmacglDPLFYD1>

When:              Monday, August 25 @ 2:00pm

 

Where:            TTI-C Conference Room: 1427 E. 60th St, 2nd Floor

 

Who:                Tong Zhang: Statistics Department, Rutgers University

 

Topic:              Approximate Solution of L0 Regularization

 

 

The problem of minimizing a convex loss function under L0 regularization is
non-smooth and non-convex. In the general case, this problem is NP hard.
However, a number of recent work showed that the problem can be
approximately solved using convex relaxation under certain assumptions. In
this talk we present a more direct approach to this problem. I will
introduce a notion of approximate local minimum for L0 regularization. It is
shown that adaptive forward-backward greedy algorithms can efficiently
compute such a local minimum.

Moreover, under appropriate assumptions, the local minimum computed by
adaptive greedy procedures approximately solves the global optimization
problem.

 

Contact:          Shai Shalev-Shwartz, TTI-C         shai@tti-c.org
834-6850

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://ttic.uchicago.edu/pipermail/colloquium/attachments/20080822/f2f7e8fa/attachment.htm