A QPTAS For Up-To-ε Revenue Maximization With Multiple Constant-Demand Bidders Over Independent Items

Dimitar Chakarov, S. Matthew Weinberg, Eric Xue

Abstract: We study revenue maximization in multi-dimensional auctions with n bidders and m items. When the bidders are constant-demand and either the number of bidders or the number of items is a constant, we give a quasi-polynomial time algorithm that computes an ε-Bayesian Incentive Compatible (ε-BIC) mechanism that obtains at least a (1 - ε) faction of the expected revenue of the optimal Bayesian Incentive Compatible (BIC) mechanism. We obtain this guarantee even when the value distribution of each bidder is unbounded, extending the main result of [Kothari et al., 2019] from a single bidder to multiple bidders.