Electrical Engineering and Computer Science


Theory Seminar

An Information Theoretical View of Information Elicitation Mechanisms

Yuqing Kong


University of Michigan
 
Friday, March 24, 2017
10:30am - 11:30am
BBB 3725

Add to Google Calendar

About the Event

Peer prediction (PP) and prediction markets (PM) are two popular and important information aggregation tools. This talk will employ information theory to provide an information theoretical understanding for both PP and PM, including recasting several important prior results into this information theory framework. This new approach promises to provide a systematic theoretical understanding for both PP and PM.

This talk will first review four important facts of Shannon information theory---(1) entropy monotonicity (2) chain rule (3) sub(super)-additivity (4) information monotonicity/data processing inequality---which are all closely related to PP and PM (and proper scoring rules, which are a key tool in both of these settings). For peer prediction, this talk will recast two important works in the PP literature---(1) Bayesian Truth Serum and (2) Dasgupta and Ghosh [2013]---into the information theoretical framework and introduce our several new PP mechanisms which are designed with the help of the information monotonicity tool. For prediction market, this talk will recast Chen and Waggoner [2016] and Aaronson [2005] into the information theoretical framework and will generalize the classical Shannon information theory concepts to provide a systematic information theoretical understanding for the prediction markets.

Additional Information

Sponsor(s): CSE

Open to: Public