Interacting Bayesian Agents on Networks

Elchanan Mossel, University of California, Berkeley
Mar 28 2013 - 4:30pm
Event type: 
PACM Colloquium
101 Sherrerd Hall

Consider a network of experts who try to predict if a certain movie will win an Oscar. When small groups of experts meet at various workshops and seminars they each state who is the more likely candidate to win in their view and then use Bayes rule to update their own private probability regarding the winner given what they've heard. This type of model has been studied extensively in theoretical economics. The main results proven in the economics literature concern the question of the convergence of opinions: do all experts end up agreeing on who is the more likely candidate to win? I will review some of the results from the economics literature and discuss a number of new results with Sly and Tamuz answering the questions of the aggregation power of this process: does this distributed process aggregate the information as well a centralized process?