1. Adoption of the Agenda - additions or deletions
2. Coffee Hour
Coffee hour last week:
Sylvain
Coffee hour this week:
???
Coffee hour next week:
???
3. Forthcoming
Date:
August 25th
September 1st
September 8th
September 15th
Location:
DC2303
DC1304
DC1304
DC1304
Chair:
Mike Watch-your-left-ski
Daming Yao
Bryan Chan
Alex Clarke
Technical
Presentation:
Mike Watch-your-left-ski
Jie Xu
Edwin Vane
Daming Yao
4. Technical Presentation
Kevin Moule
Title:COLLADA: A Brief Introduction
Abstract: COLLADA is an
interchange format for digital assest exchange. It was showcased at
this year's SIGGRAPH. I will be briefly discussing the internals of the
format, touching on some of the novel and interesting features.
5. General Discussion Items
DC1304 not available for lab meeting on Aug 25.
Please keep this until that time.
Network update: Firewall is being set up and backups are complete.
Michael Bowling: -- PMWednesday18August2004Computer
Science Dept., University of AlbertaWednesday, 18 August
200448 28 14 13 7 104 5 225 1 wlrushcsSeminarDC
2306CfalsetrueArtifical
Intelligence Group129.97.74.97wlrushMichael
BowlingLearning in a
multiagent system is a challenging problem due to two key factors.
First, if other agents are simultaneously learning then the environment
is no longer stationary, thus undermining convergence guarantees.
Second, learning is often susceptible to deception, where the other
agents may be able to exploit a learner's particular dynamics. In the
worst case, this could result in poorer performance than if the agent
was not learning at all. These challenges are identifiable in the two
most common evaluation criteria for multiagent learning algorithms:
convergence and regret. Algorithms focusing on convergence or regret in
isolation are numerous. In this talk, I seek to address both criteria in
a single algorithm by introducing GIGA-WoLF, a learning algorithm for
normal-form games. The algorithm guarantees at most zero average
regret, and converges in many situations of self-play, with both
theoretical and empirical evidence. Finally, these results also suggest
a third new learning criterion combining convergence and regret, called
subzero regret.Convergence and No-Regret in
Multiagent Learning Convergence and No-Regret in Multiagent Learning
Convergence and No-Regret in Multiagent Learning Convergence and
No-Regret in Multiagent LearningConvergence and No-Regret in Multiagent
Learning2004 7 18 14
00 0