[Intelligence Seminar] OR Seminar on August 29, 2014

[Jill Lentz]

**Distributed via the Faculty Services mail distribution system.**



The following OR seminar has been posted at:

http://server1.tepper.cmu.edu/Seminars/seminar.asp?sort=1&short=Y



Please take the time to schedule a meeting with the speaker on an individual basis.  When you are at the seminar site, proceed to the OR seminar page.  To add yourself for a meeting, click on View/Edit Schedule link and then click on the Edit Schedule link.  Enter the name portion of your e-mail address (the @andrew.cmu.edu part is not needed) and click Update at bottom of page.

-----------------------------------------------------------



Name:  Meinoff Sellmann



University:  Thomas J. Watson Research Center



Date:  August 29, 2014



Time: 1:30 to 3:00 pm



Location:  Tepper Faculty Conference Room 322



Title: Automatic Algorithm Configuration

Or: How to Win Solver Competitions Without Actually Writing a Solver



Abstract: Practically all algorithms have parameters and it is widely recognized that parameter settings often have an essential influence on algorithm performance. In this talk, we give a comprehensive overview of the state- of-the-art in algorithm portfolios and (instance-specific) algorithm tuning.



In particular, we present the robust, inherently parallel genetic algorithm GGA for the problem of configuring algorithms automatically.

GGA tunes algorithms with categorical, ordinal, and/or continuous parameters based on a training benchmark set of representative input instances.



We then consider the problem of instance-specific algorithm selection and review recent advances in algorithm portfolios which are based on non-model-based cost-sensitive multi-classification. Combining the ideas of (instance oblivious) algorithm tuning and algorithm portfolios, we finally arrive at methods for instance-specific algorithm tuning where a high-performance parameterization for a target algorithm is chosen based on the particular input.



Extensive numerical experiments evidence the effectiveness of these methods. Most notably, the methods presented in this talk have led to numerous winning entries at five international MaxSAT and SAT solver competitions.





Biography:



Meinolf Sellmann received his doctorate degree in 2002 from Paderborn University (Germany) and then went on to Cornell University as Postdoctoral Associate. From 2004 to 2010 he held a position as Assistant Professor at Brown University. He now leads a team on AI for Optimization in the cognitive computing department at IBM Watson Research.

Meinolf has published over 60 articles in international conferences and journals, served as Conference Chair of CP 2007, PC Chair of CPAIOR 2013, and associate editor of the Informs Journal on Computing. He received an NSF Early Career Award in 2007 and IBM Outstanding Technical Innovation Awards in 2013 and 2014. For four years in a row Meinolf and his team won at international SAT and MaxSAT Solver Competitions, among others two gold medals for the most CPU-time efficient SAT solver for random and crafted SAT instances in 2011, the best multi-engine approach for industrial SAT instances in 2012, the overall most efficient parallel SAT Solver in

2013 (at which point portfolios were permanently banned from the SAT competition), and seven and four first places at the 2013 and 2014 MaxSAT Evaluations.






Jill M. Lentz
Language Technologies Institute
School of Computer Science
Carnegie Mellon University
6509 Gates Hillman Complex
5000 Forbes Avenue
Pittsburgh, PA 15213
jlentz at cs.cmu.edu<mailto:jlentz at cs.cmu.edu>

T:  (412)268-1593
F:  (412)268-6298


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/intelligence-seminar-announce/attachments/20140819/0532f5ce/attachment.html>