Mal-AAEGAN generates adversarial malware in a continuous latent space
Original title: Mal-AAEGAN: Adversarial Autoencoder with Double-Critic WGAN-GP for Malware Generation
It matches or beats three binary-space GAN baselines against eight detectors on two API-feature datasets.
Researchers propose Mal-AAEGAN, an adversarial autoencoder with a double-critic WGAN-GP that perturbs malware in a continuous latent space of binary API features, then decodes while keeping original API calls. On datasets with 139 and 215 features, it is compared with MalGAN, Mal-LSGAN, and Mal-D2GAN against eight ML detectors. The paper reports stable training and evasion on par with or higher than those baselines, with the generator still responding to noise.
- Input dimensionality tested is M in {139, 215}.
- A decode-binarize-OR step preserves the malware's original API calls.
- Two critics act as a substitute detector and an additional detector.
- The generator avoids collapsing to one fixed perturbation per sample.
From the original article
1 min read at arxiv.orgMachine-learning (ML) malware detectors are known to be vulnerable to adversarial samples.
By Nam Hoang Thanh · 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.