Anti-dynamic malicious API sequence generation method based on multi-generator Auggan

Abstract

The present invention proposes a multi-generator AugGAN-based anti-dynamic malicious API sequence generation method. The present invention inserts a black box detector between the generator and the discriminator to perform unsupervised learning on sample labeling; uses multiple generators to select non-interference APIs from the corresponding non-interference API database; the selected non-interference APIs are randomly inserted into malicious APIs The sequence generates anti-dynamic malicious API samples; the adversarial samples and benign samples are input into the black box detector for classification and labeling; the adversarial samples, benign samples and their corresponding labels are input into the discriminator for training, and the gradient information is returned; the generator and the discriminator are based on The returned gradient information updates its own parameters; the continuous game between the generator and the discriminator continuously improves the generation and discrimination capabilities, and finally reaches the Nash equilibrium state; after the system is balanced, the generated anti-dynamic malicious API sequence can bypass the detection of the black box detector.

Description

technical field [0001] The invention relates to a method for generating an anti-dynamic API sequence, in particular to a method for generating an anti-dynamic malicious API sequence based on a multi-generator AugGAN. Background technique [0002] With the continuous in-depth integration and development of the new generation of information technology and work, life, and study, the Internet is penetrating into all aspects of human social activities and has become an indispensable part of people all over the world. According to incomplete statistics from the International Telecommunication Organization, by 2016, there were 3.5 billion Internet users worldwide, accounting for half of the world's total population. It is estimated that by 2020, there will be 12 billion Internet terminal devices. With the continuous emergence of new changes and characteristics of global network threats, especially some new types of network attacks emerging one after another and showing a trend of ...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to solve the problem of how to ensure deceptiveness, maliciousness and executableness of adversarial examples generated in the field of information security

Images