Protecting Bidder Information in Neural Auctions

Avi Schwarzschild, Alex Stein, Michael Curry, Tom Goldstein, John P Dickerson

Keywords: Mechanism design, neural auctions, privacy

Abstract: Single-shot auctions take place all the time, for example when selling ad space or allocating radio frequencies. Devising mechanisms for auctions with many bidders and multiple items can be complicated. It has been shown that neural networks can be used to approximate these mechanisms by satisfying the constraints that an auction be strategyproof and revenue maximizing. We show that despite such auctions maximizing revenue, they do so at the cost of revealing private bidder information. While randomness is often used to build in privacy, in this context it comes with complications if done without care. Specifically, it can violate rationality and feasibility constraints and can fundamentally change the incentive structure of the mechanism, and/or harm top-level metrics such as revenue or social welfare. We propose a method based on stochasticity that ensures privacy and meets the requirements for auction mechanisms. Furthermore, we analyze the cost to the auction house in expected revenue that comes with introducing privacy of various degrees.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: General Machine Learning (ie none of the above)

TL;DR: Neural auctions often reveal private bidder information; we apply stochasticity to make them private.