Improving Robustness of Malware Classifiers using Adversarial Strings Generated from Perturbed Latent Representations

TL;DR

This paper presents a method to enhance malware classifier robustness by generating realistic adversarial strings from perturbed latent representations, effectively defending against filename-based evasion tactics.

Contribution

It introduces an unsupervised latent space approach combined with gradient-based attacks to create adversarial examples, improving classifier resilience without sacrificing standard accuracy.

Findings

Classifiers trained on adversarial strings are more robust against filename modifications.

The method maintains high accuracy on non-adversarial data.

Significant robustness gains are achieved with minimal accuracy trade-off.

Abstract

In malware behavioral analysis, the list of accessed and created files very often indicates whether the examined file is malicious or benign. However, malware authors are trying to avoid detection by generating random filenames and/or modifying used filenames with new versions of the malware. These changes represent real-world adversarial examples. The goal of this work is to generate realistic adversarial examples and improve the classifier's robustness against these attacks. Our approach learns latent representations of input strings in an unsupervised fashion and uses gradient-based adversarial attack methods in the latent domain to generate adversarial examples in the input domain. We use these examples to improve the classifier's robustness by training on the generated adversarial set of strings. Compared to classifiers trained only on perturbed latent vectors, our approach produces classifiers that are significantly more robust without a large trade-off in standard accuracy.