Wolfram Conen, XONAR GmbH; Tuomas Sandholm, Carnegie Mellon University
Winner determination in combinatorial auctions has received significant interest in the AI community in the last 3 years. Another difficult problem in combinatorial auctions is that of eliciting the bidders’ preferences. We introduce a progressive, partial-revelation mechanism that determines an efficient allocation and the Vickrey payments. The mechanism is based on a family of algorithms that explore the natural lattice structure of the bidders’ combined preferences. The mechanism elicits utilities in a natural sequence, and aims at keeping the amount of elicited information and the effort to compute the information minimal. We present analytical results on the amount of elicitation. We show that no value-querying algorithm that is constrained to querying feasible bundles can save more elicitation than one of our algorithms. We also show that one of our algorithms can determine the Vickrey payments as a costless by-product of determining an optimal allocation.