Title: Multi-agent Deception in Attack-Defense Stochastic Game

Abstract:

In adversarial scenarios, defending oneself by using deception has recently been studied. A popular direction is to design deceptive defense strategies when the defender has complete information of the game and the attacker doesn’t. The work on deception so far models the games as a signal game or as resource allocation. In real-world scenarios where the attacker constantly learns the game by observing the movement of the defenders, the defenders could move in a way to deceive the attacker about the true game configuration. In this work, first, we introduce a new sequential simultaneous-move game model where the attacker is uncertain about part of the game configuration and attempts to learn based on its observation of defender moves. Second, we propose an algorithm to compute a k-step deceptive defense strategy to actively deceive the attacker by manipulating the attacker’s belief of the target configuration. Finally, we provide some experimental results showing that our algorithm effectively deceives the attacker and protects targets.

Zoom link: https://cmu.zoom.us/j/91660577257?pwd=RnZ3ZDZzQ25FNEYyTUV5ZCs1SmFnUT09

Meeting ID: 916 6057 7257

Password: 402632

Committee:
Katia Sycara (advisor)

Changliu Liu

Sha Yi