UTCS Colloquium/AI-Matthew Taylor/University of Southern California: "Balancing Multi-agent Exploration and Exploitation in Time-Critical Domains" TAY 3.128 Friday, May 29, 2009 11:00 a.m.

Contact Name: 
Jenna Whitney
Date: 
May 29, 2009 11:00am - 12:00pm

There is a signup schedule for this event (UT EID required).

Type of
Talk:  UTCS Colloquium/AI

Speaker/Affiliation:  Matthew
Taylor/University of Southern California

Date/Time:  Friday,

May 29, 2009  11:00 a.m.

Location:  TAY 3.128

Host

:  Peter Stone

Talk Title:  "Balancing Multi-agent

Exploration and Exploitation in Time-Critical Domains"

Talk Ast

ract:

Substantial work has investigated balancing exploration and exp

loitation, but relatively little has addressed this trade-off in the conte

xt of coordinated multi-agent interactions. In this talk I will introduce a
class of problems in which agents must maximize their on-line reward, a d

ecomposable function dependent on pairs of agents decisions. Unlike previou

s work, agents must both learn the reward function and exploit it on-line

, critical properties for a class of physically motivated systems, such as
mobile wireless networks. I will introduce algorithms motivated by the Dis

tributed Constraint Optimization Problem framework and demonstrate when, a

nd at what cost, increasing agents coordination can improve the global rew

ard on such problems.

I will also briefly discuss a couple of ad

ditional projects I have worked on at USC in the past two semesters.

Speaker Bio:

Matthew E. Taylor is a postdoctoral research associate a

t the University of Southern California working under Milind Tambe. He grad

uated magna cum laude with a double major in computer science and physics f

rom Amherst College in 2001. After working for two years as a software deve

loper, he began his Ph.D. with a MCD fellowship from the College of Natura

l Sciences. He received his doctorate from the Department of Computer Scien

ces at the University of Texas at Austin in the summer of 2008. Current res

earch interests include multi-agent systems, reinforcement learning, and

transfer learning.