Adversarial Attacks and Defenses in Machine Learning-Empowered Communication Systems and Networks: A Contemporary Survey

Wang, Yulong; Sun, Tong; Li, Shenghong; Yuan, Xin; Ni, Wei; Hossain, Ekram; Vincent Poor, H.

doi:10.1109/comst.2023.3319492

Cited by 27 publications

(6 citation statements)

References 276 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To this aim, several specific query strategies exist depending on the type of model to be attacked, for instance, support vector machines [6], or deep neural networks [7,35,36,37,38]. dataset in the quality of the replica and the effect of pruning the mimetic model (a decision tree) on its comprehensibility [40], developing also an MML-based strategy [41] to minimise the number of queries.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Cracking black-box models: Revealing hidden machine learning techniques behind their predictions

Fabra-Boluda,

Ferri,

Hernández-Orallo

et al. 2024

IDA

View full text Add to dashboard Cite

The quest for transparency in black-box models has gained significant momentum in recent years. In particular, discovering the underlying machine learning technique type (or model family) from the performance of a black-box model is a real important problem both for better understanding its behaviour and for developing strategies to attack it by exploiting the weaknesses intrinsic to the learning technique. In this paper, we tackle the challenging task of identifying which kind of machine learning model is behind the predictions when we interact with a black-box model. Our innovative method involves systematically querying a black-box model (oracle) to label an artificially generated dataset, which is then used to train different surrogate models using machine learning techniques from different families (each one trying to partially approximate the oracle’s behaviour). We present two approaches based on similarity measures, one selecting the most similar family and the other using a conveniently constructed meta-model. In both cases, we use both crisp and soft classifiers and their corresponding similarity metrics. By experimentally comparing all these methods, we gain valuable insights into the explanatory and predictive capabilities of our model family concept. This provides a deeper understanding of the black-box models and increases their transparency and interpretability, paving the way for more effective decision making.

show abstract

Section: Related Workmentioning

confidence: 99%

“…(Deep) Neural Networks [7], Naive Bayes [8], and even various online prediction APIs [9]. Knowing the model's family can thus help predict its vulnerabilities and promote more comprehensive defence strategies against adversarial attacks [10,11].…”

Section: Introductionmentioning

confidence: 99%

Cracking black-box models: Revealing hidden machine learning techniques behind their predictions

Fabra-Boluda,

Ferri,

Hernández-Orallo

et al. 2024

IDA

View full text Add to dashboard Cite

show abstract

“…Relay-based networks are increasingly utilized in device-to-device and dronecell networks for enhanced capacity and coverage, utilizing MDPs to assess performance metrics like transmission time, buffer overflow, and effective throughput [20]. The research findings in [21] demonstrate that adversarial assaults have the potential to exploit vulnerabilities in machine learning and deep neural network-based algorithms, namely in the domains of wireless signal categorization, modulation scheme detection, and MIMO network resource allocation.…”

Section: B Existing Work In Security Involving Network Slicing and Ma...mentioning

confidence: 99%

Security Challenges in 6G Networks: A Game Theory-based Intrusion Detection with Network Slicing and AI/ML Perspectives

Gupta,

Jha,

Sabraj

2024

Preprint

View full text Add to dashboard Cite

The vast network of interconnected devices of the Internet of Things poses significant security challenges, necessitating the upgrade of current wireless networks to the 6G standard and an improved intrusion detection system for host networks. Thus this article therefore underscores a game theory concept to analyze the security impediments in wireless communication systems, where a host network bandwidth is jeopardized by the intruder while acting as a legitimate user. While game theory can help with the impending security concerns, network slicing is saliently proposed for large and complicated networks, precipitating a layer wise fragmentation of the concerned network and effectively addressing the security issues in different layers of the IoT framework. However, with increasing network complexity, network security issues incorporating traditional measures become increasingly cumbersome. Therefore, this paper also briefly mentions some of the ML techniques that can be used to classify and segregate valid users from the compromised nodes in the host network. The paper discusses the potential of incorporating intelligence into wireless communication systems to enhance the host network security. The paper further aids in analyzing existing security issues in different layers of wireless communication systems, underscoring network slicing for further classification of intruders from user nodes and training the host network using ML/AI per the user application specification.

show abstract

“…Further discussions extend to offering a literature review and proposing future research directions to address existing gaps [6]. A detailed survey of adversarial attacks and defenses in ML-empowered communication systems highlights the ongoing arms race between attackers and defenders [7]. The application of ML in network anomaly detection includes various techniques and their effectiveness in identifying unusual patterns indicative of cyber threats [8].…”

Section: Introductionmentioning

confidence: 99%

Integration of Machine Learning and Human Expertise for the Improvement of Cybersecurity: Applications and Challenges

Sarker,

Das,

Hasan

2024

J. Eng. Res. Rep.

View full text Add to dashboard Cite

Cybersecurity has become an increasingly critical concern in today's digital age, with the growing sophistication of cyber threats posing significant challenges to organizations worldwide. In response to these challenges, a growing interest has been in leveraging machine learning (ML) techniques to enhance cybersecurity measures. However, while ML offers promising capabilities in detecting and mitigating cyber threats, its effectiveness can be limited when deployed in isolation. This paper explores the integration of ML algorithms with human expertise as a holistic approach to bolstering cybersecurity defenses. We discuss the applications of this integrated approach across various cybersecurity domains, highlighting its potential to improve threat detection, incident response, and vulnerability management. Additionally, we examine the challenges and considerations associated with integrating ML and human expertise, including data privacy concerns, model interpretability, and human-machine collaboration. The evaluation metrics for the Case Studies provide a detailed analysis of the effectiveness of integrating Machine Learning (ML) and Human Expertise in real-world scenarios, highlighting improvements in threat detection accuracy, incident response efficiency, and vulnerability management. The combined approach leads to more accurate and reliable threat detection, faster and more effective incident response, and enhanced operational efficiency and collaboration while maintaining high ethical and privacy standards. So, analyzing existing research and case studies, we provide insights into best practices for successfully implementing and optimizing the integration of ML and human expertise in cybersecurity operations.

show abstract

Adversarial Attacks and Defenses in Machine Learning-Empowered Communication Systems and Networks: A Contemporary Survey

Cited by 27 publications

References 276 publications

Cracking black-box models: Revealing hidden machine learning techniques behind their predictions

Cracking black-box models: Revealing hidden machine learning techniques behind their predictions

Security Challenges in 6G Networks: A Game Theory-based Intrusion Detection with Network Slicing and AI/ML Perspectives

Integration of Machine Learning and Human Expertise for the Improvement of Cybersecurity: Applications and Challenges

Contact Info

Product

Resources

About