March 2013


Options: Use Monospaced Font
Show HTML Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Debra Schenaker <[log in to unmask]>
Reply To:
Debra Schenaker <[log in to unmask]>
Sat, 23 Mar 2013 12:01:23 -0400
text/plain (3413 bytes) , text/html (47 kB)
Please mark your calendar and plan to attend ~

*C4I Seminar Series: Combinatorial Prediction Markets by Graphical 
Model: Algorithms and Auto-Traders, *by Dr. Wei Sun and Walter Powell. 
Friday, March 29, 2013 at 1:30 pm, Engineering Building, room 4705. For 
more information, contact Deb Schenaker, [log in to unmask] 
<mailto:[log in to unmask]>, ext 3-3682 or visit the Center website, 


Prediction markets are defined as speculative markets created for the 
purpose of making predictions. The current market prices can be 
interpreted as estimates of the probability of the event, or the 
expected value of the parameter. Public prediction markets such as the 
Iowa Electronic Market or the Foresight Exchange have been in place for 
over two decades.  More recently, Intrade, Inkling, and Betfair have 
been in the news.

All of these prediction markets ignore the relationships between 
questions, but /combinatorial /prediction markets explicitly consider 
and exploit dependencies among base events. This allow us to collect 
more information and promises better accuracy.  A combinatorial market 
can integrate partial information from many people, and update a joint 
probability distribution that is far larger than any one person can 
fully edit or consider. However, we must tame the combinatorial 
explosion before the problem is beyond computers as well.

In this talk, we show how to use Bayesian networks to represent and 
update combinatorial markets -- including user assets.  We also present 
results of a recent murder-mystery experiment where participants used 
either a regular prediction market or a combinatorial market to solve 
the mystery. Finally, in order to further improve the market's accuracy, 
we designed an auto-trader based on user's input and/or belief expressed 
as a Bayesian network fragment.  We show results that simple 
auto-traders can encourage participation, and new work on a "Kelly Rule" 
auto-trader that finds optimal trades given a user's joint beliefs.


*A Research Assistant Professor*in the Center of Excellence in C4I at 
George Mason University, since August 2009, Dr. Wei Sun is currently 
involved as a core researcher in a government funded research project 
called DAGGRE, which has awarded the GMU research team with more than $5 
million dollars in research funding.An expert in Bayesian inference, Dr. 
Sun obtained his Ph.D. in Information Technology in 2007 and has 
developed several efficient inference algorithms for hybrid Bayesian 
networks. He has a rich experience in predictive modeling, probabilistic 
reasoning, nonlinear filtering, sampling methods and simulation. 
Applications of his research include sensor fusion, tracking, 
classification, forecasting, performance modeling, and recently 
prediction markets. Dr. Sun has published 20 technical papers in 
referred journals and prestigious conferences, and two book chapters.

*Walter Powell is a Ph.D. candidate*and Research Instructor in George 
Mason University's C4I center.  A retired naval officer, his is 
completing his doctoral research in the evaluation of the quality of 
decisions.  As part of his research he has developed numerous 
experiments and evaluations that assessed the usefulness of various 
decisions support systems.  As a Senior Research Engineer with RTSync, 
Inc., he consults on modeling and simulation projects for various 
governmental and industry entities.