Tutorial: Computational Issues in Auction Design

 

This tutorial focuses on the interaction between computational and incentive constraints in auction design. The tutorial is designed for a computer science audience, and it should have interest to both novices and experts in computational mechanism design. No background is assumed in game theory or auction theory. We will being with a broad taxonomy of auction problems, a brief introduction of some necessary terms from game theory, and an overview of mechanism design as it applies to auction theory. The central focus of the tutorial is on four main areas of computational interest:

 

I. Centralized computational complexity.

 

We first consider the computational complexity placed on the center, in computing the outcome and payments given agent bids; special attention is given to implications on the strategic aspects of the auction. Particular topics include: (a) tractable special cases; (b) leveraging the structure in bidding languages; (c) maximal-in-range approximations; (d) decentralized and indirect implementations; (e) very fast and scalable implementations.

 

II. Communication and Valuation complexity.

 

We review the communication and preference-elicitation complexity of auction mechanisms, and consider the following computational approaches: (a) compact bidding languages (incl. interdependent valuations); (b) restricted bidding languages; (c) iterative auctions; (d) communication-and information- constrained auction design; (e) incremental revelation principles.

 

III. Strategic Complexity/ Solution concepts

 

Another problem facing agents in an auction is the strategic complexity, or the complexity of computing an equilibrium in an auction.  We review a number of methods and issues with strategic complexity, including: (a) verification vs. computation;  (b) strategyproofness and building blocks; (c) alternative solution concepts; (d) providing and exploiting structure;

 

IV. Design complexity/automated mechanism design

 

Finally, we consider the design complexity of the auction problem itself, and consider methods to automate this problem and also to fold in an explicit consideration of computational constraints into auction design. We review the basic ``design as optimization'' approach of mechanism design, outline challenges and open problems, and consider the space of automated methods to construct mechanisms, including systematic search, analysis, and evolutionary computation.

 

Biosketch of the presenter:

Dr. David Parkes is a Gordon McKay Assistant Professor of Computer Science at Harvard University, where he teaches in artificial intelligence and topics at the interface between computer science and economics. He

received his Ph.D. degree in Computer and Information Science from the University of Pennsylvania in 2001, and an M.Eng. (First class) in Engineering and Computing Science from Oxford University in 1995. Dr. Parkes received the IBM Institute of Advanced Commerce award for best dissertation proposal in electronic commerce in 2000, and is a recent recipient of an IBM Faculty Partnership award in the area of Electronic commerce. Dr. Parkes is the author of a number of patents, and serves as a technical advisor to CombineNet Inc., a company that specializes in advanced optimization software for electronic markets. For more information please visit www.eecs.harvard.edu/~parkes.