Computational Bundling for Auctions

8 years 11 days ago

Download www.cs.cmu.edu

Revenue maximization in combinatorial auctions (and other multidimensional selling settings) is one of the most important and elusive problems in mechanism design. The optimal design is unknown, and is known to include features that are not acceptable in many applications, such as favoring some bidders over others and randomization. In this paper, we instead study a common revenue-enhancement approach—bundling—in the context of the most commonly studied combinatorial auction mechanism, the Vickrey-Clarke-Groves (VCG) mechanism. A second challenge in mechanism design for combinatorial auctions is that the prior distribution on each bidder’s valuation can be doubly exponential. Such priors do not exist in most applications. Rather, in many applications (such as premium display advertising markets), there is essentially a point prior, which may not be accurate. We adopt the point prior model, and prove robustness to inaccuracy in the prior. Then, we present a branch-and-bound frame...

Christian Kroer, Tuomas Sandholm

Real-time Traffic