Time+Place: Wednesday 27/08/2014 14:30 Room 601 Taub Bld.
Title: Strict Action Uncertainty
Speaker: Reshef Meir - CSpecial Lecture - Note unusual day and place http://people.seas.harvard.edu/~rmeir/
Affiliation: Harvard School of Engineering and Applied Sciences
Host: Shaul Markovitch

Abstract:

I will introduce a natural and general model of uncertainty for games 
with many players. Intuitively, the states of the world are all action 
profiles of the agents. Each agent considers as "possible" (without 
assigning probabilities) all profiles that are close to the actual state 
according to some natural metric. Several classical approaches to 
decision making under strict uncertainty are applied, such as weak 
dominance and regret minimization.

I will focus on two applications of the model: to the analysis of 
Plurality voting (joint work with Omer Lev and Jeff Rosenschein), and 
for congestion games under uncertainty (joint work with David Parkes).

I show that the behavior of rational players under uncertainty in both 
settings follows a simple heuristics, and always converges to a pure 
equilibrium. Further, equilibria often reproduce patterns observed in 
the real world, and moderate uncertainty improves the welfare of the 
society.