Colloquium: Combinatorial optimization for national security applications

Event Type: 
Colloquium
Speaker: 
Dr. Cynthia Phillips. Sandia National Laboratories
Event Date: 
Thursday, April 13, 2017 -
3:30pm to 5:00pm
Location: 
SMLC 356
Audience: 
General Public

Event Description: 

Abstract:

National-security optimization problems emphasize safety, efficient
resource allocation, and regulation compliance.  Thus combinatorial
optimization problems from govenment frequently have some twist.  They
may be large and important enough to merit parallel solution.  They
may have to run on small, weak platforms.  They may have unusual
constraints or objectives.  Most require some degree of confidence in
the solution through testing or proofs.
 
 
We will describe several such applications.  We give an overview of a
massively-parallel branch-and-bound implementation for the
maximum-monomial-agreement problem, a core task for boosting-based
machine classification.  A challenging spam-classification problem
scales perfectly to over 6000 processors.  We will sketch
theoretically justified algorithms for a clean combinatorial
scheduling problem motivated by nuclear weapons inspections.  We will
describe water security work with the US EPA.  Finally, time
permitting, we'll mention recent work in history-independent data
structures, which give an extra level of cybersecurity.
 
Coffe and cookies will be served in the lounge at 15.00. 

Event Contact

Contact Name: Deborah Sulsky

Contact Email: sulsky@math.unm.edu