XAI to Explore Robustness of Features in Adversarial Training for Cybersecurity

Article Computer Science, Information Systems

Effective and Efficient Hybrid Android Malware Classification Using Pseudo-Label Stacked Auto-Encoder

Samaneh Mahdavifar et al.

Summary: Due to its popularity, Android has become a target for attackers. The detection of Android mobile malware is of increasing importance. Supervised machine learning is not perfect because it requires a significant amount of labeled data. Therefore, we propose a semi-supervised learning technique called pseudo-label stacked auto-encoder (PLSAE) to detect Android malware.

JOURNAL OF NETWORK AND SYSTEMS MANAGEMENT (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

ROULETTE: A neural attention multi-output model for explainable Network Intrusion Detection

Giuseppina Andresini et al.

Summary: Network Intrusion Detection (NID) systems are crucial for network protection, but existing deep learning methods are too complex to interpret. In this paper, a new neural model called ROULETTE is proposed, which combines attention mechanism and multi-output deep learning strategy for accurate and explainable classification of network traffic data. Experimental results on two benchmark datasets demonstrate the effectiveness of the proposed method in terms of accuracy and explainability.

EXPERT SYSTEMS WITH APPLICATIONS (2022)

Add to Collection

Article Computer Science, Information Systems

Adversarial EXEmples: A Survey and Experimental Evaluation of Practical Attacks on Machine Learning for Windows Malware Detection

Luca Demetrio et al.

Summary: Recent work demonstrates that adversarial Windows malware samples can bypass machine learning-based detection by manipulating a small number of input bytes. By developing a unifying framework and three novel attacks, researchers achieved better evasion rates and payload sizes, outperforming previous attacks and enabling evasion of robust models. The findings have been made open source to facilitate reproducibility and future mitigation strategies.

ACM TRANSACTIONS ON PRIVACY AND SECURITY (2021)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

Dealing with Imbalanced Data in Multi-class Network Intrusion Detection Systems Using XGBoost

Malik AL-Essa et al.

Summary: Network intrusion detection is a critical cybersecurity issue, and machine learning is considered a relevant approach. However, imbalanced data can lead to difficulties in recognizing rare attacks. Combining oversampling and feature selection can help address this issue and improve detection accuracy.

MACHINE LEARNING AND PRINCIPLES AND PRACTICE OF KNOWLEDGE DISCOVERY IN DATABASES, PT II (2021)

Add to Collection

Article Computer Science, Information Systems

LSGAN-AT: enhancing malware detector robustness against adversarial examples

Jianhua Wang et al.

Summary: This study proposes a novel approach LSGAN-AT to enhance the robustness of ML-based malware detectors against Adversarial Examples, achieving better transferability of AME in attacking 6 ML detectors and RMD in resisting the MalGAN black-box attack. The experiment results validate the effectiveness of the generated RMD in the recognition rate of AME.

CYBERSECURITY (2021)

Add to Collection

Proceedings Paper Computer Science, Information Systems

Troubleshooting an Intrusion Detection Dataset: the CICIDS2017 Case Study

Gints Engelen et al.

Summary: The paper explores the effectiveness of machine learning in network intrusion detection and the challenges faced in applying it to large-scale network environments. By revisiting the CICIDS2017 dataset and addressing issues in data processing, improvements in model evaluation were achieved. Addressing data collection issues can have a significant impact on the performance of machine learning algorithms and recommendations for anticipation and prevention are provided.

2021 IEEE SYMPOSIUM ON SECURITY AND PRIVACY WORKSHOPS (SPW 2021) (2021)

Add to Collection

Article Computer Science, Theory & Methods

Adversarial XAI Methods in Cybersecurity

Aditya Kuppa et al.

Summary: Machine Learning is crucial in cybersecurity, with explanation methods shedding light on black-box classifiers. Recent research focuses on improving explainability, attacking interpreters, and defining properties of explanations. However, there is a lack of thorough study on how model explanations can introduce new attack surfaces, with potential privacy-compromising attacks.

IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY (2021)

Add to Collection

Article Computer Science, Hardware & Architecture