Skip to main content

Amir Khazraei

Duke University, Electrical and Computer Engineering, Graduate Student

University of Tehran, School of Electrical and Computer Engineering, Graduate Student

Followers

3

Following

2

Co-author

1

Public Views

I received my B.Sc degree in Electrical Engineering in 2014 from Amirkabir university and my M.Sc degree in 2017 from university of Tehran. Currently, I am a PhD candidate at Duke university.

less

University of Pennsylvania

University of Pennsylvania

New York University

sunandan adhikary

Toru Namerikawa

Keio University

Qatar University

Abel Cadenillas

Interests

Uploads

Papers by Amir Khazraei

Learning-Based Vulnerability Analysis of Cyber-Physical Systems

ArXiv, 2021

This work focuses on the use of deep learning for vulnerability analysis of cyber-physical system... more This work focuses on the use of deep learning for vulnerability analysis of cyber-physical systems (CPS). Specifically, we consider a control architecture widely used in CPS (e.g., robotics), where the low-level control is based on e.g., the extended Kalman filter (EKF) and an anomaly detector. To facilitate analyzing the impact potential sensing attacks could have, our objective is to develop learning-enabled attack generators capable of designing stealthy attacks that maximally degrade system operation. We show how such problem can be cast within a learning-based grey-box framework where parts of the runtime information are known to the attacker, and introduce two models based on feed-forward neural networks (FNN); both models are trained offline, using a cost function that combines the attack effects on the estimation error and the residual signal used for anomaly detection, so that the trained models are capable of recursively generating such effective sensor attacks in real-tim...

Resiliency of Perception-Based Controllers Against Attacks

This work focuses on resiliency of learning-enabled perception-based controllers for nonlinear dy... more This work focuses on resiliency of learning-enabled perception-based controllers for nonlinear dynamical systems. We consider systems equipped with an end-to-end controller, mapping the perception (e.g., camera images) and sensor measurements to control inputs, as well as a statistical or learning-based anomaly detector (AD). We define a general notion of attack stealthiness and find conditions for which there exists a sequence of stealthy attacks on perception and sensor measurements that forces the system into unsafe operation without being detected, for any employed AD. Specifically, we show that systems with unstable physical plants and exponentially stable closedloop dynamics are vulnerable to such stealthy attacks. Finally, we use our results on a case-study.

Learning-Based Vulnerability Analysis of Cyber-Physical Systems

This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems... more This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems (CPS). Specifically, we consider a control architecture widely used in CPS (e.g., in robotics), where the low-level control is based on e.g., the extended Kalman filter (EKF) and an anomaly detector. To facilitate analyzing the impact potential sensing attacks could have, our objective is to develop learning-enabled attack generators capable of designing stealthy attacks that maximally degrade system operation. We show how such problem can be cast within a learning-based grey-box framework where only parts of the runtime information are known to the attacker, and introduce two models based on feed-forward neural networks (FNN) and recurrent neural networks (RNN). Both models are trained offline, using a cost function that combines the attack impact on the estimation error (and thus control) and the residual signal used for anomaly detection, so that the trained models are capable of recursively generating effective yet stealthy sensor attacks in real-time. The effectiveness of the proposed methods is illustrated on several case studies.

Replay attack detection in a multi agent system using stability analysis and loss effective watermarking

2017 American Control Conference (ACC), 2017

In this paper the problem of detecting replay attack in a homogeneous multi agent system is addre... more In this paper the problem of detecting replay attack in a homogeneous multi agent system is addressed. Information exchange among the agents is under network topology and each agent is controlled to satisfy both local and global objectives. A local estimator and an anomaly detector are designed for each agent. Under replay attack to a part of network, we provide conditions on the stability of the under attack system in which, the attack cannot be detected by none of the agents. Next, a detection method is proposed which is based on using a watermarked control strategy, in which watermarking signal is shared among the agents through the network. Nevertheless, since this exogenous watermarking signal affects the performance of the controlled system, an optimization problem is introduced to find a watermarking signal with minimum loss effect.

An Optimal Linear Dynamic Detection Method for Replay Attack in Cyber-Physical Systems

The problem of detecting replay attack to the linear stochastic system with Kalman filer state es... more The problem of detecting replay attack to the linear stochastic system with Kalman filer state estimator and LQG controller is addressed. To this end, a dynamic attack detector method is proposed which is coupled with the dynamics of the system. While preserving stability of the main system, conditions on parameters of the attack detector dynamics are obtained such that the attack can be revealed by destabilization of a residual trajectory which is the difference between the estimated and measured output of the system. Using this method, system operator can adjust the detection rate based on the proposed scheme by changing the design parameters. Nevertheless, since the exogenous attack detector signal affects the performance of the closed loop control system, we propose an optimization problem to determine such a detector with minimum loss effect. In the simulation results, the proposed dynamical attack detector approach is compared with the well-known additive white noise watermark...

Attack-resilient state estimation with intermittent data authentication

Automatica, 2021

Network-based attacks on control systems may alter sensor data delivered to the controller, effec... more Network-based attacks on control systems may alter sensor data delivered to the controller, effectively causing degradation in control performance. As a result, having access to accurate state estimates, even in the presence of attacks on sensor measurements, is of critical importance. In this paper, we analyze performance of resilient state estimators (RSEs) when any subset of sensors may be compromised by a stealthy attacker. Specifically, we consider systems with the well-known l0-based RSE and two commonly used sound intrusion detectors (IDs). For linear time-invariant plants with bounded noise, we define the notion of perfect attackability (PA) when attacks may result in unbounded estimation errors while remaining undetected by the employed ID (i.e., stealthy). We derive necessary and sufficient PA conditions, showing that a system can be perfectly attackable even if the plant is stable. While PA can be prevented with the use the standard cryptographic mechanisms (e.g., message authentication) that ensure data integrity under network-based attacks, their continuous use imposes significant communication and computational overhead. Consequently, we also study the impact that even intermittent use of data authentication has on RSE performance guarantees in the presence of stealthy attacks. We show that if messages from some of the sensors are even intermittently authenticated, stealthy attacks could not result in unbounded state estimation errors.

Perfect Attackability of Linear Dynamical Systems with Bounded Noise

2020 American Control Conference (ACC), 2020

This paper addresses the problem of secure state estimation in the presence of attacks on sensor ... more This paper addresses the problem of secure state estimation in the presence of attacks on sensor measurements of a linear time invariant (LTI) system. We assume that the system is equipped with a common l0-based attack-resilient state estimator and a sound anomaly detector. We introduce the notion of perfect attackability (PA) for LTI systems with bounded noise, when the attacker may introduce an unbounded estimation error while remaining undetected by the anomaly detector. Finally, necessary and sufficient conditions for perfectly attackable systems are provided, and illustrated on examples.

A new watermarking approach for replay attack detection in LQG systems

2017 IEEE 56th Annual Conference on Decision and Control (CDC), 2017

The durability of a building is determined to a great extent by the quality of materials used in ... more The durability of a building is determined to a great extent by the quality of materials used in its construction. Sandcrete block is a major building material in Nigeria. This research assessed the quality of blocks produced in Adeta, Ilorin West Local Government Area of Kwara State, Nigeria. Eight block factories were selected from the study area and the quality of their blocks were measured for their water absorption ratios and compressive strengths and compared with the Nigerian Industrial Standards (NIS) specifications. Six samples each of size 450mm x 225mm x 225mm per factory were tested. Experimental findings revealed that the absorption rate of the sandcrete blocks were higher than the 12% specification. The compressive strengths of the blocks which ranges from 0.19N/mm 2 to 0.40N/mm 2 fall below the minimum specification of 2.5N/mm 2. Inadequate mix ratio is a factor that resulted in the poor quality sandcrete blocks produced in the area.

Attack-Resilient StateEstimationwith IntermittentData Authentication ?

Network-based attacks on control systems may alter sensor data delivered to the controller, effec... more Network-based attacks on control systems may alter sensor data delivered to the controller, effectively causing degradation in control performance. As a result, having access to accurate state estimates, even in the presence of attacks on sensor measurements, is of critical importance. In this paper, we analyze performance of resilient state estimators (RSEs) when any subset of sensors may be compromised by a stealthy attacker. Specifically, we consider systems with the well-known l0-based RSE and two commonly used sound intrusion detectors (IDs). For linear time-invariant plants with bounded noise, we define the notion of perfect attackability (PA) when attacks may result in unbounded estimation errors while remaining undetected by the employed ID (i.e., stealthy). We derive necessary and sufficient PA conditions, showing that a system can be perfectly attackable even if the plant is stable. While PA can be prevented with the use the standard cryptographic mechanisms (e.g., message...

replay attack detection in a multi agent system using stability analysis and loss effective watermarking

Drafts by Amir Khazraei

Learning-Based Vulnerability Analysis of Cyber-Physical Systems

by Amir Khazraei and Spencer Hallyburton

This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems... more This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems (CPS). Specifically, we consider a control architecture widely used in CPS (e.g., in robotics), where the low-level control is based on e.g., the extended Kalman filter (EKF) and an anomaly detector. To facilitate analyzing the impact potential sensing attacks could have, our objective is to develop learning-enabled attack generators capable of designing stealthy attacks that maximally degrade system operation. We show how such problem can be cast within a learning-based grey-box framework where only parts of the runtime information are known to the attacker, and introduce two models based on feed-forward neural networks (FNN) and recurrent neural networks (RNN). Both models are trained offline, using a cost function that combines the attack impact on the estimation error (and thus control) and the residual signal used for anomaly detection, so that the trained models are capable of recursively generating effective yet stealthy sensor attacks in real-time. The effectiveness of the proposed methods is illustrated on several case studies.

Learning-Based Vulnerability Analysis of Cyber-Physical Systems

ArXiv, 2021

This work focuses on the use of deep learning for vulnerability analysis of cyber-physical system... more This work focuses on the use of deep learning for vulnerability analysis of cyber-physical systems (CPS). Specifically, we consider a control architecture widely used in CPS (e.g., robotics), where the low-level control is based on e.g., the extended Kalman filter (EKF) and an anomaly detector. To facilitate analyzing the impact potential sensing attacks could have, our objective is to develop learning-enabled attack generators capable of designing stealthy attacks that maximally degrade system operation. We show how such problem can be cast within a learning-based grey-box framework where parts of the runtime information are known to the attacker, and introduce two models based on feed-forward neural networks (FNN); both models are trained offline, using a cost function that combines the attack effects on the estimation error and the residual signal used for anomaly detection, so that the trained models are capable of recursively generating such effective sensor attacks in real-tim...

Resiliency of Perception-Based Controllers Against Attacks

This work focuses on resiliency of learning-enabled perception-based controllers for nonlinear dy... more This work focuses on resiliency of learning-enabled perception-based controllers for nonlinear dynamical systems. We consider systems equipped with an end-to-end controller, mapping the perception (e.g., camera images) and sensor measurements to control inputs, as well as a statistical or learning-based anomaly detector (AD). We define a general notion of attack stealthiness and find conditions for which there exists a sequence of stealthy attacks on perception and sensor measurements that forces the system into unsafe operation without being detected, for any employed AD. Specifically, we show that systems with unstable physical plants and exponentially stable closedloop dynamics are vulnerable to such stealthy attacks. Finally, we use our results on a case-study.

Learning-Based Vulnerability Analysis of Cyber-Physical Systems

This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems... more This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems (CPS). Specifically, we consider a control architecture widely used in CPS (e.g., in robotics), where the low-level control is based on e.g., the extended Kalman filter (EKF) and an anomaly detector. To facilitate analyzing the impact potential sensing attacks could have, our objective is to develop learning-enabled attack generators capable of designing stealthy attacks that maximally degrade system operation. We show how such problem can be cast within a learning-based grey-box framework where only parts of the runtime information are known to the attacker, and introduce two models based on feed-forward neural networks (FNN) and recurrent neural networks (RNN). Both models are trained offline, using a cost function that combines the attack impact on the estimation error (and thus control) and the residual signal used for anomaly detection, so that the trained models are capable of recursively generating effective yet stealthy sensor attacks in real-time. The effectiveness of the proposed methods is illustrated on several case studies.

Replay attack detection in a multi agent system using stability analysis and loss effective watermarking

2017 American Control Conference (ACC), 2017

In this paper the problem of detecting replay attack in a homogeneous multi agent system is addre... more In this paper the problem of detecting replay attack in a homogeneous multi agent system is addressed. Information exchange among the agents is under network topology and each agent is controlled to satisfy both local and global objectives. A local estimator and an anomaly detector are designed for each agent. Under replay attack to a part of network, we provide conditions on the stability of the under attack system in which, the attack cannot be detected by none of the agents. Next, a detection method is proposed which is based on using a watermarked control strategy, in which watermarking signal is shared among the agents through the network. Nevertheless, since this exogenous watermarking signal affects the performance of the controlled system, an optimization problem is introduced to find a watermarking signal with minimum loss effect.

An Optimal Linear Dynamic Detection Method for Replay Attack in Cyber-Physical Systems

The problem of detecting replay attack to the linear stochastic system with Kalman filer state es... more The problem of detecting replay attack to the linear stochastic system with Kalman filer state estimator and LQG controller is addressed. To this end, a dynamic attack detector method is proposed which is coupled with the dynamics of the system. While preserving stability of the main system, conditions on parameters of the attack detector dynamics are obtained such that the attack can be revealed by destabilization of a residual trajectory which is the difference between the estimated and measured output of the system. Using this method, system operator can adjust the detection rate based on the proposed scheme by changing the design parameters. Nevertheless, since the exogenous attack detector signal affects the performance of the closed loop control system, we propose an optimization problem to determine such a detector with minimum loss effect. In the simulation results, the proposed dynamical attack detector approach is compared with the well-known additive white noise watermark...

Attack-resilient state estimation with intermittent data authentication

Automatica, 2021

Network-based attacks on control systems may alter sensor data delivered to the controller, effec... more Network-based attacks on control systems may alter sensor data delivered to the controller, effectively causing degradation in control performance. As a result, having access to accurate state estimates, even in the presence of attacks on sensor measurements, is of critical importance. In this paper, we analyze performance of resilient state estimators (RSEs) when any subset of sensors may be compromised by a stealthy attacker. Specifically, we consider systems with the well-known l0-based RSE and two commonly used sound intrusion detectors (IDs). For linear time-invariant plants with bounded noise, we define the notion of perfect attackability (PA) when attacks may result in unbounded estimation errors while remaining undetected by the employed ID (i.e., stealthy). We derive necessary and sufficient PA conditions, showing that a system can be perfectly attackable even if the plant is stable. While PA can be prevented with the use the standard cryptographic mechanisms (e.g., message authentication) that ensure data integrity under network-based attacks, their continuous use imposes significant communication and computational overhead. Consequently, we also study the impact that even intermittent use of data authentication has on RSE performance guarantees in the presence of stealthy attacks. We show that if messages from some of the sensors are even intermittently authenticated, stealthy attacks could not result in unbounded state estimation errors.

Perfect Attackability of Linear Dynamical Systems with Bounded Noise

2020 American Control Conference (ACC), 2020

This paper addresses the problem of secure state estimation in the presence of attacks on sensor ... more This paper addresses the problem of secure state estimation in the presence of attacks on sensor measurements of a linear time invariant (LTI) system. We assume that the system is equipped with a common l0-based attack-resilient state estimator and a sound anomaly detector. We introduce the notion of perfect attackability (PA) for LTI systems with bounded noise, when the attacker may introduce an unbounded estimation error while remaining undetected by the anomaly detector. Finally, necessary and sufficient conditions for perfectly attackable systems are provided, and illustrated on examples.

A new watermarking approach for replay attack detection in LQG systems

2017 IEEE 56th Annual Conference on Decision and Control (CDC), 2017

The durability of a building is determined to a great extent by the quality of materials used in ... more The durability of a building is determined to a great extent by the quality of materials used in its construction. Sandcrete block is a major building material in Nigeria. This research assessed the quality of blocks produced in Adeta, Ilorin West Local Government Area of Kwara State, Nigeria. Eight block factories were selected from the study area and the quality of their blocks were measured for their water absorption ratios and compressive strengths and compared with the Nigerian Industrial Standards (NIS) specifications. Six samples each of size 450mm x 225mm x 225mm per factory were tested. Experimental findings revealed that the absorption rate of the sandcrete blocks were higher than the 12% specification. The compressive strengths of the blocks which ranges from 0.19N/mm 2 to 0.40N/mm 2 fall below the minimum specification of 2.5N/mm 2. Inadequate mix ratio is a factor that resulted in the poor quality sandcrete blocks produced in the area.

Attack-Resilient StateEstimationwith IntermittentData Authentication ?

Network-based attacks on control systems may alter sensor data delivered to the controller, effec... more Network-based attacks on control systems may alter sensor data delivered to the controller, effectively causing degradation in control performance. As a result, having access to accurate state estimates, even in the presence of attacks on sensor measurements, is of critical importance. In this paper, we analyze performance of resilient state estimators (RSEs) when any subset of sensors may be compromised by a stealthy attacker. Specifically, we consider systems with the well-known l0-based RSE and two commonly used sound intrusion detectors (IDs). For linear time-invariant plants with bounded noise, we define the notion of perfect attackability (PA) when attacks may result in unbounded estimation errors while remaining undetected by the employed ID (i.e., stealthy). We derive necessary and sufficient PA conditions, showing that a system can be perfectly attackable even if the plant is stable. While PA can be prevented with the use the standard cryptographic mechanisms (e.g., message...

replay attack detection in a multi agent system using stability analysis and loss effective watermarking

Learning-Based Vulnerability Analysis of Cyber-Physical Systems

by Amir Khazraei and Spencer Hallyburton

This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems... more This work focuses on the use of deep learning for vulnerability analysis of cyberphysical systems (CPS). Specifically, we consider a control architecture widely used in CPS (e.g., in robotics), where the low-level control is based on e.g., the extended Kalman filter (EKF) and an anomaly detector. To facilitate analyzing the impact potential sensing attacks could have, our objective is to develop learning-enabled attack generators capable of designing stealthy attacks that maximally degrade system operation. We show how such problem can be cast within a learning-based grey-box framework where only parts of the runtime information are known to the attacker, and introduce two models based on feed-forward neural networks (FNN) and recurrent neural networks (RNN). Both models are trained offline, using a cost function that combines the attack impact on the estimation error (and thus control) and the residual signal used for anomaly detection, so that the trained models are capable of recursively generating effective yet stealthy sensor attacks in real-time. The effectiveness of the proposed methods is illustrated on several case studies.