Using adversarial examples to deceive end-to-end deep learning malware detectors
Deceiving End-to-End Deep Learning Malware Detectors using Adversarial Examples
AI summary
Researchers proposed a new loss function for discrete inputs such as executable bytes, using adversarial examples to deceive end-to-end deep learning malware detectors. The method injects a small byte payload into malicious binaries so that they are judged as benign by a convolutional deep learning detection model while retaining their original functionality, achieving a high rate of detection evasion. The payload transfers across different positions within the same file and across different files, and has low entropy, similar to benign data segments.