Deep False-Name-Proof Auction Mechanisms

We explore an approach to designing false-name-proof auction mechanisms using deep learning. While multi-agent systems researchers have recently proposed data-driven approaches to automatically designing auction mechanisms through deep learning, false-name-proofness, which generalizes strategy-proofness by assuming that a bidder can submit multiple bids under fictitious identifiers, has not been taken into account as a property that a mechanism has to satisfy. We extend the RegretNet neural network architecture to incorporate false-name-proof constraints and then conduct experiments demonstrating that the generated mechanisms satisfy false-name-proofness.