Malicious software confrontation sample generation method combined with API fuzzy processing technology

Abstract

The invention relates to a malicious software confrontation sample generation method combined with an API fuzzy processing technology, and belongs to the technical field of computers and information science. The method comprises the following steps: firstly, inserting irrelevant API vectors into an original API calling sequence to preliminarily realize obfuscation of malicious software feature functions; then performing confusion operation of function call redirection on the sample, and hiding function functions of an API; and finally, feeding the modified sample to a malware classifier, checking whether the malware classifier realizes error classification or not, and solving an optimization problem by utilizing a genetic algorithm. The invention has higher practicability and better meetsthe actual requirements; compared with a black-box-oriented adversarial sample generation method, the invention has the advantages that original functions of malicious software are not damaged, the success rate is high when the malware classifier is cheated, and the problems that the success rate of cheating adversarial samples is low and the functions of the malicious software are damaged are greatly solved.

Description

technical field [0001] The invention relates to a method for generating malicious software confrontation samples combined with API fuzzy processing technology, and belongs to the technical field of computer and information science. Background technique [0002] Due to a large number of malware attacks in cyberspace, machine learning techniques have been widely used in malware detection and classification. In order to evaluate the anti-attack performance of malware detection models, it is particularly important to study attack methods for malware detection models. At the same time, the deep model itself has been proved to be vulnerable to the attack of adversarial samples. So far, although adversarial learning has been an active research field, most of the research on adversarial samples has been applied in the field of image recognition. Research based on deep learning Adversarial attack methods for malware detection systems are in the minority, and most adversarial sample g...

AI Technical Summary

Problems solved by technology

[0003] The basic problems to be solved in malware countermeasure generation methods are: generating irrelevant APIs to insert into the original API sequence, hiding imported API functions through an API obfuscation technique, and deceiving malware classifiers into misclassifications with a high success rate

But these methods usually break the functionality of the malware, resulting in the inability to create adversarial examples that can work effectively in practice.

[0008] To sum up, among the existing malware adversarial sample generation methods, the black-box-oriented attack method is more feasible, but it is difficult to guarantee that the malware function will not be destroyed after the disturbance is added, and that the adversarial samples can be used in real-world applications. counter attack

Images