Skip to main navigation Skip to search Skip to main content

Statistical Privacy-Preserving Online Nash Equilibrium Learning with Two-Point Bandit Feedback

  • Shuoshuo ZHANG*
  • , Meng LUAN
  • , Dan ZHAO
  • *Corresponding author for this work

Research output: Book Chapters | Papers in Conference ProceedingsConference paper (refereed)Researchpeer-review

Abstract

This paper investigates distributed online Nash equilibrium learning with privacy preservation in dynamic non-cooperative games. In this context, players optimize their actions under local constraints to track the time-varying equilibrium without access to future cost information while protecting their private strategy information. To this end, a novel privacy-preserving distributed online algorithm with two-point bandit feedback is proposed, which enables Nash equilibrium tracking using only local function evaluations. Moreover, theoretical analysis establishes sublinear dynamic regret under sublinear path variation. Finally, numerical simulations validate theoretical results.
Original languageEnglish
Title of host publicationProceedings of 2025 IEEE International Conference on Unmanned Systems, ICUS 2025
EditorsRong SONG
PublisherIEEE
Pages1568-1573
Number of pages6
ISBN (Electronic)9798331526726
ISBN (Print)9798331526733
DOIs
Publication statusPublished - 2025
Externally publishedYes

Bibliographical note

Publisher Copyright:
© 2025 IEEE.

Funding

This work was supported in part by the National Key Research and Development Program of China under Grant 2022YFA1004702, in part by the National Natural Science Foundation of China through Grant 62325304, Grant U22B2046, Grant 62073079, and Grant 62088101, and in part by the Jiangsu Provincial Scientific Research Center of Applied Mathematics under Grant BK20233002.

Keywords

  • Nash equilibrium seeking
  • bandit feedback
  • privacy preservation
  • decentralized online game

Fingerprint

Dive into the research topics of 'Statistical Privacy-Preserving Online Nash Equilibrium Learning with Two-Point Bandit Feedback'. Together they form a unique fingerprint.

Cite this