"Auction Learning as a Two Player Game": GANs (?) for Mechanism Design

Anonymous

Keywords: mechanism design, deep learning, auctions, game theory

Abstract: Designing strategyproof, revenue-maximizing auctions is an important task, but it is surprisingly difficult -- even in some seemingly trivial cases, nothing is known about the optimal auction design. Motivated by this lack of progress, a number of recent works have proposed the use of deep neural networks as function approximators for learning strategyproof mechanisms. One of these works is "Auction Learning as a Two-Player Game", which appeared at ICLR 2021. We discuss this work, situate it in the broader context of deep learning for auctions, explain how it improves over prior techniques, and discuss the future outlook for interactions between modern deep learning and mechanism design.

ICLR Paper: https://openreview.net/forum?id=YHdeAO61l6T