### Abstract

This article addresses the problem of designing optimal strategies in consensus protocols for networks vulnerable to adversarial attacks. First, a set of necessary conditions for optimal control is given in the case of the dynamic (multi-stage) weight selection problem of consensus protocols. Under some mild conditions, it turns out that only one-stage is sufficient for reaching consensus, and the article derives a closed-form solution for the optimal control. Second, a (zero-sum) game theoretical model with a “convex-convex” quadratic objective function is considered for the problem of a network with an adversary corrupting the control signal with noise. Mixed-strategy saddle-point (MSSP) strategies are obtained for the players (the adversary and the network designer) in the resulting game. Further, a totally distributed gradient method that computes the optimal control is provided. Simulation results show that an adversary using an MSSP strategy can drive the system away from consensus, while an adversary using a uniform random strategy does not cause as much damage.

Original language | English (US) |
---|---|

Title of host publication | Annals of the International Society of Dynamic Games |

Publisher | Birkhauser |

Pages | 117-137 |

Number of pages | 21 |

DOIs | |

State | Published - Jan 1 2016 |

### Publication series

Name | Annals of the International Society of Dynamic Games |
---|---|

Volume | 14 |

ISSN (Print) | 2474-0179 |

ISSN (Electronic) | 2474-0187 |

### Fingerprint

### Keywords

- Adversary
- Consensus protocols
- Convex-convex zero-sum quadratic games
- Distributed control
- Saddle-point strategies

### ASJC Scopus subject areas

- Statistics, Probability and Uncertainty
- Statistics and Probability
- Applied Mathematics

### Cite this

*Annals of the International Society of Dynamic Games*(pp. 117-137). (Annals of the International Society of Dynamic Games; Vol. 14). Birkhauser. https://doi.org/10.1007/978-3-319-28014-1_6

**A zero-sum game between the network designer and an adversary in consensus protocols.** / El Chamie, Mahmoud; Başar, Tamer.

Research output: Chapter in Book/Report/Conference proceeding › Chapter

*Annals of the International Society of Dynamic Games.*Annals of the International Society of Dynamic Games, vol. 14, Birkhauser, pp. 117-137. https://doi.org/10.1007/978-3-319-28014-1_6

}

TY - CHAP

T1 - A zero-sum game between the network designer and an adversary in consensus protocols

AU - El Chamie, Mahmoud

AU - Başar, Tamer

PY - 2016/1/1

Y1 - 2016/1/1

N2 - This article addresses the problem of designing optimal strategies in consensus protocols for networks vulnerable to adversarial attacks. First, a set of necessary conditions for optimal control is given in the case of the dynamic (multi-stage) weight selection problem of consensus protocols. Under some mild conditions, it turns out that only one-stage is sufficient for reaching consensus, and the article derives a closed-form solution for the optimal control. Second, a (zero-sum) game theoretical model with a “convex-convex” quadratic objective function is considered for the problem of a network with an adversary corrupting the control signal with noise. Mixed-strategy saddle-point (MSSP) strategies are obtained for the players (the adversary and the network designer) in the resulting game. Further, a totally distributed gradient method that computes the optimal control is provided. Simulation results show that an adversary using an MSSP strategy can drive the system away from consensus, while an adversary using a uniform random strategy does not cause as much damage.

AB - This article addresses the problem of designing optimal strategies in consensus protocols for networks vulnerable to adversarial attacks. First, a set of necessary conditions for optimal control is given in the case of the dynamic (multi-stage) weight selection problem of consensus protocols. Under some mild conditions, it turns out that only one-stage is sufficient for reaching consensus, and the article derives a closed-form solution for the optimal control. Second, a (zero-sum) game theoretical model with a “convex-convex” quadratic objective function is considered for the problem of a network with an adversary corrupting the control signal with noise. Mixed-strategy saddle-point (MSSP) strategies are obtained for the players (the adversary and the network designer) in the resulting game. Further, a totally distributed gradient method that computes the optimal control is provided. Simulation results show that an adversary using an MSSP strategy can drive the system away from consensus, while an adversary using a uniform random strategy does not cause as much damage.

KW - Adversary

KW - Consensus protocols

KW - Convex-convex zero-sum quadratic games

KW - Distributed control

KW - Saddle-point strategies

UR - http://www.scopus.com/inward/record.url?scp=85031779310&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85031779310&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-28014-1_6

DO - 10.1007/978-3-319-28014-1_6

M3 - Chapter

AN - SCOPUS:85031779310

T3 - Annals of the International Society of Dynamic Games

SP - 117

EP - 137

BT - Annals of the International Society of Dynamic Games

PB - Birkhauser

ER -