Cyber Physical Systems (CPS) are characterized by their ability to integrate the physical and information or cyber worlds. Their deployment in critical infrastructure have demonstrated a potential to transform the world. However, harnessing this potential is limited by their critical nature and the far reaching effects of cyber attacks on human, infrastructure and the environment. An attraction for cyber concerns in CPS rises from the process of sending information from sensors to actuators over the wireless communication medium, thereby widening the attack surface. Traditionally, CPS security has been investigated from the perspective of preventing intruders from gaining access to the system using cryptography and other access control techniques. Most research work have therefore focused on the detection of attacks in CPS. However, in a world of increasing adversaries, it is becoming more difficult to totally prevent CPS from adversarial attacks, hence the need to focus on making CPS resilient. Resilient CPS are designed to withstand disruptions and remain functional despite the operation of adversaries. One of the dominant methodologies explored for building resilient CPS is dependent on machine learning (ML) algorithms. However, rising from recent research in adversarial ML, we posit that ML algorithms for securing CPS must themselves be resilient. This paper is therefore aimed at comprehensively surveying the interactions between resilient CPS using ML and resilient ML when applied in CPS. The paper concludes with a number of research trends and promising future research directions. Furthermore, with this paper, readers can have a thorough understanding of recent advances on ML-based security and securing ML for CPS and countermeasures, as well as research trends in this active research area.

Resilient Machine Learning for Networked Cyber Physical Systems: A Survey for Machine Learning Security to Securing Machine Learning for CPS

Cyber Physical Systems (CPS) are characterized by their ability to integrate the physical and information or cyber worlds. Their deployment in critical infrastructure have demonstrated a potential to transform the world. However, harnessing this potential is limited by their critical nature and the far reaching effects of cyber attacks on human, infrastructure and the environment. An attraction for cyber concerns in CPS rises from the process of sending information from sensors to actuators over the wireless communication medium, thereby widening the attack surface. Traditionally, CPS security has been investigated from the perspective of preventing intruders from gaining access to the system using cryptography and other access control techniques. Most research work have therefore focused on the detection of attacks in CPS. However, in a world of increasing adversaries, it is becoming more difficult to totally prevent CPS from adversarial attacks, hence the need to focus on making CPS resilient. Resilient CPS are designed to withstand disruptions and remain functional despite the operation of adversaries. One of the dominant methodologies explored for building resilient CPS is dependent on machine learning (ML) algorithms. However, rising from recent research in adversarial ML, we posit that ML algorithms for securing CPS must themselves be resilient. This article is therefore aimed at comprehensively surveying the interactions between resilient CPS using ML and resilient ML when applied in CPS. The paper concludes with a number of research trends and promising future research directions. Furthermore, with this article, readers can have a thorough understanding of recent advances on ML-based security and securing ML for CPS and countermeasures, as well as research trends in this active research area.

Biofeatures are fast becoming a key tool to authenticate the IoT devices; in this sense, the purpose of this investigation is to summarise the factors that hinder biometrics models’ development and deployment on a large scale, including human physiological (e.g., face, eyes, fingerprints-palm, or electrocardiogram) and behavioral features (e.g., signature, voice, gait, or keystroke). The different machine learning and data mining methods used by authentication and authorization schemes for mobile IoT devices are provided. Threat models and countermeasures used by biometrics-based authentication schemes for mobile IoT devices are also presented. More specifically, we analyze the state of the art of the existing biometric-based authentication schemes for IoT devices. Based on the current taxonomy, we conclude our paper with different types of challenges for future research efforts in biometrics-based authentication schemes for IoT devices.

/pdf/authentication-and-authorization-for-mobile-iot-devices-589r3orxhd.pdf

Authentication and Authorization for Mobile IoT Devices Using Biofeatures: Recent Advances and Future Trends

兪 焄著 「政府企業關係論」(서울 : 法文社, 1989)

Research has produced many types of authentication systems that use machine learning. However, there is no consistent approach for reporting performance metrics and the reported metrics are inadequate. In this work, we show that several of the common metrics used for reporting performance, such as maximum accuracy (ACC), equal error rate (EER) and area under the ROC curve (AUROC), are inherently flawed. These common metrics hide the details of the inherent tradeoffs a system must make when implemented. Our findings show that current metrics give no insight into how system performance degrades outside the ideal conditions in which they were designed. We argue that adequate performance reporting must be provided to enable meaningful evaluation and that current, commonly used approaches fail in this regard. We present the unnormalized frequency count of scores (FCS) to demonstrate the mathematical underpinnings that lead to these failures and show how they can be avoided. The FCS can be used to augment the performance reporting to enable comparison across systems in a visual way. When reported with the Receiver Operating Characteristics curve (ROC), these two metrics provide a solution to the limitations of currently reported metrics. Finally, we show how to use the FCS and ROC metrics to evaluate and compare different authentication systems.

/pdf/robust-performance-metrics-for-authentication-systems-3fh28vxkaq.pdf

Robust Performance Metrics for Authentication Systems.

Recent research demonstrated that the superficially well-trained machine learning (ML) models are highly vulnerable to adversarial examples. As ML techniques are becoming a popular solution for cyber-physical systems (CPSs) applications in research literatures, the security of these applications is of concern. However, current studies on adversarial machine learning (AML) mainly focus on pure cyberspace domains. The risks the adversarial examples can bring to the CPS applications have not been well investigated. In particular, due to the distributed property of data sources and the inherent physical constraints imposed by CPSs, the widely-used threat models and the state-of-the-art AML algorithms in previous cyberspace research become infeasible. 
We study the potential vulnerabilities of ML applied in CPSs by proposing Constrained Adversarial Machine Learning (ConAML), which generates adversarial examples that satisfy the intrinsic constraints of the physical systems. We first summarize the difference between AML in CPSs and AML in existing cyberspace systems and propose a general threat model for ConAML. We then design a best-effort search algorithm to iteratively generate adversarial examples with linear physical constraints. We evaluate our algorithms with simulations of two typical CPSs, the power grids and the water treatment system. The results show that our ConAML algorithms can effectively generate adversarial examples which significantly decrease the performance of the ML models even under practical constraints.

ConAML: Constrained Adversarial Machine Learning for Cyber-Physical Systems

Motivated by the great potential of implicit and seamless user authentication, we attempt to build an implicit authentication (IA) system with adaptive sampling that automatically selects dynamic sets of activities for user behavior extraction. Various activities, such as user location, application usage, user motion, and battery usage have been popular choices to generate behaviors, the soft biometrics, for implicit authentication. Unlike password-based or hard biometric-based authentication, implicit authentication does not require explicit user action or expensive hardware. However, user behaviors can change unpredictably which renders it more challenging to develop systems that depend on them. In addition to dynamic behavior extraction, the proposed implicit authentication system differs from the existing systems in terms of energy efficiency for battery-powered mobile devices. Since implicit authentication systems including the proposed one rely on machine learning, the expensive training process needs be outsourced to the remote server. However, mobile devices may not always have reliable network connections to send real-time data to the server for training. We overcome this limitation by proposing a W-layer, an overlay that provides a practical and energy-efficient solution for implicit authentication on mobile devices. We implemented partially labeled Dirichlet allocation (PLDA) on the server side for more accurate feature extraction, and achieved 93.3 percent precision and 98.6 percent accuracy in the synthetic dataset. Furthermore, we tested the power consumption of the smartphones used for our experiments and found that our method consumed 14.5 percent of the devices’ total battery usage.

PersonaIA: A Lightweight Implicit Authentication System Based on Customized User Behavior Selection

Motivated by the great potential of implicit and seamless user authentication, we attempt to build an efficient middle layer running on mobile devices to support implicit authentication (IA) systems with adaptive sampling. Various activities, such as user location, application usage, user motion, and battery usage have been popular choices to generate behaviors, the soft biometrics, for implicit authentication. Unlike password-based or hard biometric-based authentication, implicit authentication does not require explicit user action or expensive hardware. However, user behaviors can change unpredictably which renders it more challenging to develop systems that depend on them. Various machine learning algorithms have been used to address this challenge. The expensive training process is usually outsourced to the remote server but this can potentially increase the chance of data leakage. In addition, mobile devices may not always have reliable network connections to send real-time data to the server for training. Motivated by these limitations, we propose a W-layer, an overlay that provides an energy-efficient solution for real-time implicit authentication on mobile devices. The size of the data the system needs to collect at different times depends on the legitimacy of the user. This in turn affects how the sampling rate is adjusted which can reduce energy consumption. To evaluate our method, we conducted several experiments on both synthetic and real datasets. The average accuracy of identifying legitimate users is 96.73% using the synthetic dataset and 96.70% using the real dataset. Furthermore, we tested the power consumption on a low-end Nexus S smartphone to obtain a more pessimistic result. We found that our method consumed 14.5% of the device's total battery usage. The power consumption performance is expected to improve significantly on high-end mobile devices.

Energy-efficient W-layer for behavior-based implicit authentication on mobile devices

Energy theft causes large economic losses to utility companies around the world. In recent years, energy theft detection approaches based on machine learning (ML) techniques, especially neural networks, are becoming popular in the research community and shown to achieve state-of-the-art detection performance. However, in this work, we demonstrate that the well-trained ML models for energy theft detection are highly vulnerable to adversarial attacks. In particular, we design an adversarial measurement generation approach that enables the attacker to report extremely low power consumption measurements to utilities while bypassing the ML energy theft detection. We evaluate our approach with three kinds of neural networks based on a real-world smart meter dataset. The evaluation results demonstrate that our approach is able to significantly decrease the ML models’ detection accuracy, even for black-box attackers.

SearchFromFree: Adversarial Measurements for Machine Learning-based Energy Theft Detection

With the rapid growth of the smart device market, associated security issues become more threatening and diverse than ever before. Due to the limitations of the traditional explicit authentication mechanisms (e.g., Password-based, biometrics), researchers and the industry have been promoting implicit authentication (IA) that does not require explicit user action and potentially enhances user experience to further protect devices from misuse. IA typically leverages various types of behavioral data to deduce a user behavior model for authentication purpose. However, IA systems are still at their infancy and exhibit many limitations, one of which is how to determine the best retraining frequency when updating the user behavior model. Another limitation is how to gracefully degrade user privilege, when authentication fails to identify legitimate users (i.e., False negatives) for a practical IA system. To address the first problem, we propose an algorithm that utilizes Jensen-Shannon (JS) dis (tance) to determine the optimal retraining frequency. For the second problem, we introduce a dynamic privilege mechanism, again based on JS-dis (tance), to achieve multi-level fine-grained access control. Our simulation results show that the proposed techniques can successfully detect the degradation of accuracy of the user behavior model, as well as automatically determine and adjust to the best retraining frequency. It is also shown that the dynamic privilege-based access control reduces the impact of false negatives on legitimate users and enhances system reliability and user experience compared with the traditional lock-only method in case of authentication failure.

Yingyuan Yang

Papers

ConAML: Constrained Adversarial Machine Learning for Cyber-Physical Systems

PersonaIA: A Lightweight Implicit Authentication System Based on Customized User Behavior Selection

Energy-efficient W-layer for behavior-based implicit authentication on mobile devices

SearchFromFree: Adversarial Measurements for Machine Learning-based Energy Theft Detection

Retraining and Dynamic Privilege for Implicit Authentication Systems