We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.GT

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Science and Game Theory

Title: Towards Prior-Free Approximately Truthful One-Shot Auction Learning via Differential Privacy

Abstract: Designing truthful, revenue maximizing auctions is a core problem of auction design. Multi-item settings have long been elusive. Recent work (arXiv:1706.03459) introduces effective deep learning techniques to find such auctions for the prior-dependent setting, in which distributions about bidder preferences are known. One remaining problem is to obtain priors in a way that excludes the possibility of manipulating the resulting auctions. Using techniques from differential privacy for the construction of approximately truthful mechanisms, we modify the RegretNet approach to be applicable to the prior-free setting. In this more general setting, no distributional information is assumed, but we trade this property for worse performance. We present preliminary empirical results and qualitative analysis for this work in progress.
Subjects: Computer Science and Game Theory (cs.GT); Machine Learning (cs.LG)
Cite as: arXiv:2104.00159 [cs.GT]
  (or arXiv:2104.00159v1 [cs.GT] for this version)

Submission history

From: Daniel Reusche [view email]
[v1] Wed, 31 Mar 2021 23:22:55 GMT (202kb)

Link back to: arXiv, form interface, contact.