[Intelligence Seminar] Tepper Operations Research Seminar, Dr. Ashish Sabharwal, Oct. 18

Dana Houston dhouston at cs.cmu.edu
Thu Oct 10 11:24:37 EDT 2013

Possibly of interest to the members of this mailing list:

The following Tepper Operations Research seminar has been posted at:


Please take the time to schedule a meeting with the speaker on an 
individual basis.When you are at the seminar site, proceed to this 
seminar announcement.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: Dr. Ashish Sabharwal

Affiliation: IBM Watson Research Center

Date: October 18, 2013

Time: 1:30 to 3:00 pm

Location: Room 153 Posner Hall

Title:Combinatorial Search in Parallel: Advances and Fundamental Barriers

ABSTRACT:With modern day hardware providing easy access to over 10,000 
cores, effective parallelization of algorithms has become a topic of 
wide interest. While approaches such as Map-Reduce have enjoyed 
tremendous success in tackling Data Parallelism, effective 
parallelization of sophisticated Combinatorial Search and Optimization 
engines remains a challenge. For instance, despite many attempts,
parallel MIP solvers drastically lose efficiency after about 16 cores 
and parallel SAT solvers in the 2011 SAT Competition achieved a speedup 
of roughly 3 when given 32 cores. This talk will discuss three threads 
we have pursued at IBM Research to address the parallelization 
challenge. First, building upon the success of data-driven Portfolio 
Methods, we will see how column-generation based parallel scheduling can 
draw the best out of diverse, non-information-sharing solvers. Second, 
we will discuss frameworks SatX10 and DDX10 that make it easier for 
researchers to design and experiment with large scale parallelization 
schemes without any expertise in low level mechanisms such as MPI. 
Third, we will see recent evidence that the proof mechanisms underlying 
complete sequential search algorithms can impose significant bottlenecks 
to their efficient parallelization. Our findings suggest that the design 
of effective parallel solvers will require new techniques that 
fundamentally alter the structure of the underlying infeasibility or 
optimality proof.---------- BIO: Ashish Sabharwal is a research 
scientist in the Business Analytics and Math Sciences (BAMS) department 
at IBM Watson Research Center. He specializes in the areas of 
Combinatorial Reasoning, Probabilistic Inference, and Discrete 
Optimization and has been developing the next generation of advanced 
analytics capabilities as part of IBM's Smarter Energy Research 
Institute (SERI). Prior to late 2010, Ashish was a Research Associate at 
Cornell University's Institute for Computational Sustainability (ICS) 
and a Postdoc at Cornell's Intelligent Information Systems Institute 
(IISI). He received his M.S. and Ph.D. from the University of 
Washington, Seattle in 2001/2005. Ashish has co-authored over 60 
research publications in major journals and conferences, in addition to 
several surveys for books and collections. He has co-developed solvers 
that have won in international SAT Competitions (2011, 2012, 2013). His 
research findings have been recognized with Best Paper Awards (AAAI, 
CP), a Best Student Paper Award (UAI), and Runner-up Prizes (UAI, 
IJCAI-JAIR 5-Year Best) during 2006-2013, in addition to several other 
nominations.---------- WEB: 

Dana M. Houston
Language Technologies Institute
School of Computer Science
Carnegie Mellon University
6511 Gates Hillman Complex
5000 Forbes Avenue
Pittsburgh, PA 15213

T:  (412)268-4717
F:  (412)268-6298

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/intelligence-seminar-announce/attachments/20131010/9573b939/attachment-0001.html>

More information about the intelligence-seminar-announce mailing list