Q-MALDetect study measures evasion of a quantum malware detector
Original title: Q-MALDetect: Quantum Image-Based Malware Detection and Adversarial Robustness Analysis
The hybrid detector reached 97.2% training accuracy, but the strongest feature-space attack succeeded 78.7% of the time.
A paper introduces Q-MALDetect, a hybrid convolutional and variational-quantum detector for byte images of x86-64 ELF binaries, and tests it against feature-space and Grad-CAM-guided executable-aware evasion. Robustness falls as attack strength rises. Dynamic adversarial training was the strongest defense evaluated, while convolutional-autoencoder preprocessing raised executable-aware attack success. Qubit count did not relate monotonically to robustness.
- Grad-CAM-guided NOP insertion had the highest executable-aware conditional ASR, 32.9%.
- Adversarial training cut feature-space ASR to 5.0% and executable-aware ASR to 7.7%.
- Sequential-attack ASR remained 34.0% under that defense.
- Autoencoder preprocessing raised executable-aware ASR to 39.7%.
- Grad-CAM mass in the mutable .text region rose from 35.2% to 45.7%.
From the original article
1 min read at arxiv.orgQuantum machine-learning (QML) models are increasingly considered for security applications, yet their adversarial robustness in malware analysis remains insufficiently understood.
By Walid El Maouaki · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.