The workshop deadline for submission has been pushed back
one week due to several requests. If those who have already
submitted would like to make revisions, they may resubmit
their paper before the new deadline.
Call for Papers:
ICML Workshop on Constrained Optimization and Structured
Output Spaces
-----------------------
A large amount of machine learning research in recent years
has focused on domains with structured output spaces, such
as protein secondary structure prediction, natural language
parsing, and various applications in computer vision and
information retrieval. Graphical models provide powerful
tools to utilize structure in these domains. In addition,
recent work has focused on the use of standard supervised
learning techniques to solve these types of problems.
At the same time, mathematical programming in machine
learning continues to be a focus of intense research. New
applications and uses for support vector machines continue
to be discovered. Integer linear programming has worked its
way into many interesting learning and inference algorithms.
New techniques for approximate constrained optimization have
opened up the possibility of solving optimizations with
exponential numbers of constraints.
To discuss the increasing cross-pollination of these two
areas, ICML 2007 is hosting a workshop on Constrained
Optimization and Structured Output Spaces. For this
workshop we invite submission of papers on original research
in the areas of constrained optimization and learning in
structured output spaces. Submissions that address the
following questions are particularly sought after:
Overview of Titular Areas: There is a great deal of work on
the intersection of these two areas. What are good examples
of the state of the art in these areas? On what
domains/types of problems has this convergence been
successful? On what types of problems has it failed and
why? When it does fail, what other methods succeed?
Generality of Techniques: Some of the applications used in
the literature require a great deal of specialization of the
candidate optimization technique. To what extent can these
specialized techniques be generalized to solve other,
seemingly unrelated problems? On what types of problems
will the candidate technique fail even if it can be applied
readily?
Heuristic vs. Exact Optimization: Many of the techniques
used for exact constrained optimization are still extremely
time consuming and often impractical. When can heuristic
approaches be used to give approximate solutions to these
optimizations and when are they appropriate? What heuristic
techniques are available? What is the trade-off between
heuristic solutions and exact ones in a given domain?
Applications: What are some large-scale, real-world
applications on which these techniques have been applied
effectively? What are some as yet untested applications,
preferably with publicly available datasets, that show
promise?
Theory: What theoretical tools are available to categorize
these problems and analyze the solution techniques? How do
the different approaches to solving these problems relate to
each other?
Submissions
-----------------------
Submissions should be no longer than six pages in length,
and should follow the ICML submission style guidelines.
Identifying information including names of authors,
affiliations, and contact information should appear on the
first page of the paper. Authors of accepted papers will be
invited to give a presentation of the work at the workshop
and final papers will be published electronically on the
workshop web site.
Please go to:
http://web.engr.orst.edu/~parker/icmlworkshop
for more information or to submit a paper.
Important Dates
------------------------
Submissions Due: 30 April 2007 (Deadline Extended)
Notification of Acceptance: 14 May 2007
Camera-ready Papers Due: 28 May 2007
Workshop Date: 24 June 2007
Organizing Committee
-----------------------
Thomas Dietterich - Oregon State University
Charles Parker - Oregon State University
Eric Xing - Carnegie Mellon University
Scott Yih - Microsoft Research
Program Committee
-----------------------
Ulf Brefeld, Humboldt-Universitt zu Berlin
Thomas Dietterich, Oregon State University
Lise Getoor, University of Maryland
Rong Jin, Michigan State University
Thorsten Joachims, Cornell University
Charles Parker, Oregon State University
Dan Roth, University of Illinois, Champaign-Urbana
Prasad Tadepalli, Oregon State University
Eric Xing, Carnegie Mellon University
Scott Yih, Microsoft Research
