Towards an Iterated Game Model with Multiple Adversaries in Smart-World Systems

Xiaofei He, Xinyu Yang, Wei Yu, Jie Lin, Qingyu YangSensors 2018 · OpenReview

TL;DR

This paper introduces a novel game-theoretic model to analyze coalitional cyber-attacks in smart-world systems, moving beyond single-adversary scenarios. By extending the Iterated Public Goods Game (IPGG) and incorporating a defender with a penalty mechanism, the authors demonstrate how strategic defense can significantly reduce adversary payoffs and mitigate attack impact.

What Problem Does This Paper Address?

Smart-world systems, encompassing smart cities, grids, and transportation, are increasingly vulnerable to sophisticated cyber-attacks. While much research focuses on defending against individual threats, a critical gap exists in understanding and defending against coalitional attacks, where multiple adversaries cooperate. These attacks pose a greater threat due to their coordinated nature. Existing defense strategies often lack a comprehensive framework to quantify the interaction between multiple attackers and a defender, particularly concerning the cost-effectiveness of defensive measures. This paper aims to fill this gap by modeling the complex dynamics of multi-adversary attacks and exploring how a defender can strategically influence adversary behavior through penalties.

Methods

The core of this research is a game-theoretic model built upon the extended Iterated Public Goods Game (IPGG). In each round, adversaries can choose to cooperate in a coalitional attack or defect. If the attack is successful, the gain is shared. Crucially, if the attack is detected by the defender, participating adversaries are penalized. The defender's capability is represented by a penalty factor ($\beta$).

The authors employ zero-determinant strategies, a class of strategies that allow a player to unilaterally control the expected payoff of their opponents. Specifically, they analyze the equalizer strategy, where a participant aims to set their opponents' payoff to a fixed value. This is extended to consider collusive strategies where multiple adversaries coordinate to control payoffs. Finally, they introduce and analyze an extortion strategy, where a participant aims to maximize their own payoff by extorting others, even at the expense of overall gains. Theoretical analysis is combined with simulations to evaluate the effectiveness of these strategies and the defender's role.

Datasets

Not discussed in the paper. The evaluation is based on simulations with parameters set according to the theoretical analysis.

Key Insights and Ideas

  • Coalitional Attacks are a Major Threat: The paper highlights that focusing solely on individual attacks is insufficient, as coordinated multi-adversary attacks are a significant and growing concern.
  • Defender's Penalty is Key: The defender's ability to impose penalties ($\beta$) is a powerful tool. By strategically setting this penalty, the defender can significantly influence adversary behavior and reduce their incentives to attack.
  • Zero-Determinant Strategies Offer Control: The use of zero-determinant strategies, particularly the equalizer strategy, allows adversaries to manipulate their opponents' payoffs, leading to complex interactions.
  • Extortion Creates Competition: The extortion strategy, while seemingly selfish, can paradoxically lead to increased competition among adversaries, making cooperation against the defender more difficult and reducing overall attack impact.
  • Balancing Defense and Cost: The model provides a framework for the defender to determine an "adequate penalty factor" to mitigate attacks, implicitly linking defense effectiveness to cost.

Key Results

The paper presents theoretical analyses and simulation results demonstrating:

  • WSLS vs. Random Strategy: The Win-Stay-Lose-Shift (WSLS) strategy generally outperforms the random strategy, gaining more payoff. Increasing the penalty factor ($\beta$) reduces the payoffs for both strategies.
  • Equalizer Strategy:
    • When adversaries use the equalizer strategy, they can set their opponents' payoffs to a fixed value.
  • If all participants adopt the equalizer strategy and aim for global optimal outcomes, they can achieve higher payoffs, especially when the penalty factor is low.
  • As the penalty factor ($\beta$) increases, the payoffs for all strategies (including equalizer) decrease.
  • Adaptive Equalizer Strategy: This strategy allows participants to achieve similar payoffs to WSLS, and when all participants use it, payoffs converge to a maximum value (when $\beta=0$). Increasing $\beta$ reduces payoffs.
  • Extortion Strategy: This strategy consistently outperforms WSLS and random strategies. It leads to increased competition among adversaries, reducing their collective payoff and making cooperation against the defender harder. The defender can leverage this by increasing $\beta$ to encourage extortion.

Limitations and Future Work

  • Simplified Adversary Models: The paper primarily focuses on specific strategies (random, WSLS, equalizer, collusive, extortion). Real-world adversaries might employ more dynamic and adaptive strategies. The authors suggest exploring adaptation strategies where adversaries learn and adjust their behavior based on opponent actions and game history.
  • Scope of Attacks: The current model focuses on a generalized IPGG framework. Future work could extend the model to analyze specific attack types (e.g., manipulating electricity prices, disrupting energy generation) and explore different game models like the Stackelberg model for more nuanced scenarios.
  • Zero-Determinant Strategy Constraints: The effectiveness of zero-determinant strategies can be limited by the number of participants or attack gain rates. The authors suggest developing new mechanisms to overcome these limitations, potentially by grouping participants or analyzing group-level interactions.
  • Nash Equilibrium Analysis: While a Nash equilibrium is shown to exist for selfish participants, further exploration of equilibria under various strategy combinations and defender actions would be valuable.

Summary updated

Abstract

Diverse and varied cyber-attacks challenge the operation of the smart-world system that is supported by Internet-of-Things (IoT) (smart cities, smart grid, smart transportation, etc.) and must be carefully and thoughtfully addressed before widespread adoption of the smart-world system can be fully realized. Although a number of research efforts have been devoted to defending against these threats, a majority of existing schemes focus on the development of a specific defensive strategy to deal with specific, often singular threats. In this paper, we address the issue of coalitional attacks, which can be launched by multiple adversaries cooperatively against the smart-world system such as smart cities. Particularly, we propose a game-theory based model to capture the interaction among multiple adversaries, and quantify the capacity of the defender based on the extended Iterated Public Goods Game (IPGG) model. In the formalized game model, in each round of the attack, a participant can either cooperate by participating in the coalitional attack, or defect by standing aside. In our work, we consider the generic defensive strategy that has a probability to detect the coalitional attack. When the coalitional attack is detected, all participating adversaries are penalized. The expected payoff of each participant is derived through the equalizer strategy that provides participants with competitive benefits. The multiple adversaries with the collusive strategy are also considered. Via a combination of theoretical analysis and experimentation, our results show that no matter which strategies the adversaries choose (random strategy, win-stay-lose-shift strategy, or even the adaptive equalizer strategy), our formalized game model is capable of enabling the defender to greatly reduce the maximum value of the expected average payoff to the adversaries via provisioning sufficient defensive resources, which is reflected by setting a proper penalty factor against the adversaries. In addition, we extend our game model and analyze the extortion strategy, which can enable one participant to obtain more payoff by extorting his/her opponents. The evaluation results show that the defender can combat this strategy by encouraging competition among the adversaries, and significantly suppress the total payoff of the adversaries via setting the proper penalty factor.
BibTeX
@inproceedings{he2018towards,
  author = {Xiaofei He and Xinyu Yang and Wei Yu and Jie Lin and Qingyu Yang},
  title = {{Towards an Iterated Game Model with Multiple Adversaries in Smart-World Systems}},
  booktitle = {Sensors},
  year = {2018},
  url = {https://openreview.net/forum?id=8T5aRSykoN},
}
Figure