Effectiveness of Adversarial Examples and Defenses for Malware\n Classification

Artificial neural networks have been successfully used for many different\nclassification tasks including malware detection and distinguishing between\nmalicious and non-malicious programs. Although artificial neural networks\nperform very well on these tasks, they are also vulnerable to adversarial\nexamples. An adversarial example is a sample that has minor modifications made\nto it so that the neural network misclassifies it. Many techniques have been\nproposed, both for crafting adversarial examples and for hardening neural\nnetworks against them. Most previous work has been done in the image domain.\nSome of the attacks have been adopted to work in the malware domain which\ntypically deals with binary feature vectors. In order to better understand the\nspace of adversarial examples in malware classification, we study different\napproaches of crafting adversarial examples and defense techniques in the\nmalware domain and compare their effectiveness on multiple datasets.\n