A Game Theoretical vulnerability analysis of Adversarial Attack

TL;DR

This paper introduces a game theoretical framework to analyze vulnerabilities of CAPTCHA-based classifiers against adversarial attacks, using real attack methods and strategic modeling to understand potential defenses.

Contribution

It presents a novel game theory approach to model and analyze adversarial attacks on deep learning classifiers in cybersecurity, specifically for CAPTCHA systems.

Findings

Game theoretical model reveals potential attack and defense strategies.

Application of FGSM and One Pixel Attack demonstrates real-world attack scenarios.

Stackelberg game analysis helps identify optimal defense strategies.

Abstract

In recent times deep learning has been widely used for automating various security tasks in Cyber Domains. However, adversaries manipulate data in many situations and diminish the deployed deep learning model's accuracy. One notable example is fooling CAPTCHA data to access the CAPTCHA-based Classifier leading to the critical system being vulnerable to cybersecurity attacks. To alleviate this, we propose a computational framework of game theory to analyze the CAPTCHA-based Classifier's vulnerability, strategy, and outcomes by forming a simultaneous two-player game. We apply the Fast Gradient Symbol Method (FGSM) and One Pixel Attack on CAPTCHA Data to imitate real-life scenarios of possible cyber-attack. Subsequently, to interpret this scenario from a Game theoretical perspective, we represent the interaction in the Stackelberg Game in Kuhn tree to study players' possible behaviors and actions by applying our Classifier's actual predicted values. Thus, we interpret potential attacks in deep learning applications while representing viable defense strategies in the game theory prospect.