[Mar. 26th] Seminar by Dr. Roger Mailler (SRI International)

• Speaker: Dr. Roger Mailler
• Location: CH 430
• Date: Wednesday, March 26
• Time: 11:00 am - 12:00 pm

Seminar Abstract

Dynamic, Partial Centralization is a method for solving distributed problems
that works by dynamically and incrementally centralizing relevant portions
of a shared problem in order to rapidly converge on a globally acceptable
solution.  This technique is effective because it uses the structure of the
problem and the current solution state to identify critical
interdependencies that can then be resolved using fast, centralized
algorithms. Philosophically, this work represents a new paradigm in
distributed problem solving because, unlike techniques that attempt to
maintain complete distribution of knowledge and control, this technique
capitalizes on the observation that using focused centralization can often
lead to faster, more efficient solutions.

In this talk, I will discuss the origin of dynamic, partial centralization
by describing a protocol that was used to solve dynamic, resource allocation
problems in a distributed sensor network.  I will show how the key ideas
from this protocol led to the creation of an entire family of algorithms
that have been used to solve both static and dynamic, distributed problems.
   

Speaker Biography

Roger Mailler received a B.S. with Honors from Stony Brook University in
1999 and a Ph.D. in Computer Science from the University of Massachusetts in
2004.  He is currently a Senior Computer Scientist for the Artificial
Intelligence Center at SRI International.  In addition to being a pioneer in
the area of dynamic, partial centralization, Dr. Mailler is also an acting
Principal Investigator on DARPA's Bootstrapped Learning program.  His
current research interests include multi-agent systems, distributed
constraint satisfaction and optimization, distributed sensor networks, and
more recently, machine learning.

Attachments