Prediction Markets and Learning from Crowds
Jake Abernethy (University of Pennsylvania)
NICTA SML SEMINARDATE: 2011-11-24
TIME: 11:00:00 - 12:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
Prediction Markets have been proposed as a tool for utilizing the crowd to receive accurate predictions about the likelihood of some uncertain event. By allowing any individual to place bets on potential outcomes, the market prices tend to reflect a type of "consensus estimate" of the true probabilities. The goal of this talk will be to take this market-based approach and apply it to more complex problems. For example, what can we do if the state space is combinatorial or even infinite? More broadly, can we design a market to train a hypothesis in a learning problem? These will be some of the ideas we explore in this talk.
Joint work with Rafael Frongillo, as well as Yiling Chen and Jenn
Wortmann Vaughn.
Anyone curious to read more on this topic:
1. Forthcoming NIPS paper: http://arxiv.org/abs/1111.2664
2. EC 2011 paper:http://www.eecs.berkeley.edu/~jake/ec041-abernethy.pdf
