Student research opportunities

Markets, Risk, Learning, and Prediction.

Project Code: CECS_161

This project is available at the following levels:
Honours, Summer Scholar
Please note that this project is only for undergraduate students.

Keywords:

decision theory, learning theory, prediction markets, machine learning

Supervisor:

Dr Mark Reid

Outline:

Decision theory is a mathematical formulation of how we can make the best possible choices in the face of uncertainty. Its applications are broad ranging—machine learning, economics, operations research, robotics—as are the techniques it makes use of—probability, statistics, game theory, geometry.

Recently, there have been a number of very interesting connections made between these different perspectives: prediction markets can be understood in terms of learning theory [CV10]; the notion of generalised information and its relationship to risk minimisation and divergences [RW09]; an information geometric understanding of losses [DL05]; and the study of losses for probability elicitation [LPS08].

By putting many of these developments into a common setting we hope to better understand the key principles of decision theory and exploit any applications they may have in machine learning.

Goals of this project

To understand some recent developments in decision theory – particularly their application to mechanism design – and their implications for the theory and practice of machine learning.

Requirements/Prerequisites

A strong background in mathematics and/or statistics. Some knowledge of convex analysis would be helpful but not necessary.

Student Gain

Students will gain a familiarity with some recent advances in learning theory, prediction markets, and their connections.

Background Literature

[RW09]: "Information, Divergence and Risk for Binary Experiments", Reid and Williamson, arXiv 0901.0356, 2009.

[CV10]: "A New Understanding of Prediction Markets Via No-Regret Learning", Chen and Vaughan, In Proc. of AISTATS, 2010.

[DL05]: "The Geometry of Decision Theory", Dawid and Lauritzen, In Proc. of Info. Geometry and its Applications, 2005.

[LPS08]: "Eliciting Properties of Probability Distributions", Lambert, Pennock and Shoham, In Proc. of the ACM Conference on Electronic Commerce, 2008.