Inception: Efficiently Computable Misinformation Attacks on Markov Games

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: McMahan, Jeremy, Wu, Young, Chen, Yudong, Zhu, Xiaojin, Xie, Qiaomin
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866907956667219968
author McMahan, Jeremy
Wu, Young
Chen, Yudong
Zhu, Xiaojin
Xie, Qiaomin
author_facet McMahan, Jeremy
Wu, Young
Chen, Yudong
Zhu, Xiaojin
Xie, Qiaomin
contents We study security threats to Markov games due to information asymmetry and misinformation. We consider an attacker player who can spread misinformation about its reward function to influence the robust victim player's behavior. Given a fixed fake reward function, we derive the victim's policy under worst-case rationality and present polynomial-time algorithms to compute the attacker's optimal worst-case policy based on linear programming and backward induction. Then, we provide an efficient inception ("planting an idea in someone's mind") attack algorithm to find the optimal fake reward function within a restricted set of reward functions with dominant strategies. Importantly, our methods exploit the universal assumption of rationality to compute attacks efficiently. Thus, our work exposes a security vulnerability arising from standard game assumptions under misinformation.
format Preprint
id arxiv_https___arxiv_org_abs_2406_17114
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Inception: Efficiently Computable Misinformation Attacks on Markov Games
McMahan, Jeremy
Wu, Young
Chen, Yudong
Zhu, Xiaojin
Xie, Qiaomin
Machine Learning
Cryptography and Security
Computer Science and Game Theory
We study security threats to Markov games due to information asymmetry and misinformation. We consider an attacker player who can spread misinformation about its reward function to influence the robust victim player's behavior. Given a fixed fake reward function, we derive the victim's policy under worst-case rationality and present polynomial-time algorithms to compute the attacker's optimal worst-case policy based on linear programming and backward induction. Then, we provide an efficient inception ("planting an idea in someone's mind") attack algorithm to find the optimal fake reward function within a restricted set of reward functions with dominant strategies. Importantly, our methods exploit the universal assumption of rationality to compute attacks efficiently. Thus, our work exposes a security vulnerability arising from standard game assumptions under misinformation.
title Inception: Efficiently Computable Misinformation Attacks on Markov Games
topic Machine Learning
Cryptography and Security
Computer Science and Game Theory
url https://arxiv.org/abs/2406.17114