Using Conventions for Learning Anti-Coordination
|
Speaker (s):
Prof. Boi FALTINGS
Director,
Artificial Intelligence Laboratory,
École Polytechnique Fédérale de Lausanne (EPFL)
|
|
Date:
Time:
Venue:
|
|
July 10, 2018, Tuesday
2:00pm - 3:00pm
Meeting Room 4.4, Level 4
School of Information Systems
Singapore Management University
80 Stamford Road
Singapore 178902
|
|
ABSTRACT
When multiple agents access a common resource, efficient equilibria require that a single agent access the resource and are thus very unfair. Symmetric equilibria are fair but inefficient, but can be made more efficient using correlation devices. We propose the Price of Anonymity as a measure to compare fair mechanisms, and show how distributed learning algorithms can achieve good efficiency even without a central authority. We propose an algorithm based on conventions to achieve fast convergence and show how it can be implemented as a simple learning rule. Finally, we consider how participation in such learning can be made rational.
About the Speaker
I obtained my undergraduate degree at the Swiss Federal Institute of Technology, Zurich (ETH) in 1983 and a Ph.D. degree from the University of Illinois, Urbana, in 1987. From 1987 to 1993, I have been professeur extraordinaire at EPFL. Since 1993, I am full professor and have served as head of the computer science department from 1996-1998. In 1997, I co-founded Iconomic Systems SA, and served as chief scientist until it was sold to i:FAO in 2001. I continued as chief scientist of i:FAO in 2001 and 2002. In 2004, I co-founded NexThink SA and in 2008, I co-founded Prediggo SA, both spinoffs from the Artificial Intelligence Laboratory. From 2004-2008, I was head of the Institute for Core Computing Sciences at EPFL. From 2007 onwards, I am the computer science member of the executive committee of the Nanotera program. From 2013-2016, I was director of the EPFL Social Media Lab. From 2014-2017, I am member of the AAAI executive council.
