Geometric vector-based adversarial sample generation method

Abstract

The invention discloses a geometric vector-based adversarial sample generation method. The adversarial sample generation method comprises the steps of 1, data preprocessing; step 2, model pre-training; 3, repeating the steps (a)-(f) until convergence to obtain a DGA domain name confrontation sample: (a) inputting a legal domain name into the ATN to generate a legal domain name confrontation sample, and obtaining disturbance loss; (b) inputting the legal domain name and the legal domain name adversarial sample into a noise disturbance direction function to obtain noise; (c) inputting the noiseand the DGA domain name into a disturbance network to obtain a DGA domain name adversarial sample, wherein the disturbance network is a disturbance network based on a geometric vector; (d) inputting the DGA domain name adversarial sample into a target network to obtain target network loss; (e) obtaining a target loss function by using the disturbance loss and the target network loss; and (f) updating the ATN network by minimizing the target loss function. According to the invention, the adversarial sample can be generated for a specific DGA category.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a geometric vector-based confrontation sample generation method. Background technique [0002] Deep neural networks are excellent at handling complex tasks, but recent research has shown that they are vulnerable to adversarial attacks, which are attacks in the form of adding tiny perturbations to the input, causing the model to predict the wrong output. In practical applications, adversarial attacks pose a serious threat to the success of deep learning. For this reason, the researchers proposed a method of adversarial sample generation to deal with potential attacks and enhance the robustness and generalization ability of neural networks. [0003] At present, in the principle of adversarial sample generation, it is mainly divided into two categories, one is to randomly add noise to the original sample until it can deceive the neural network, and the other is in the dire...

AI Technical Summary

Problems solved by technology

However, these methods have the following problems: not any perturbation of the original sample will cause classification errors, only specific perturbations will cause classification errors, so by adding noise randomly, it is not easy to get adversarial samples

Although the gradient information gives the perturbation direction, for high-dimensional data and more complex neural network models, it will increase the complexity of establishing gradient information

Existing adversarial sample generation methods do not take into account the cost and efficiency of generating samples

Images