Anti-attack detection method and system based on network node topological structure

Abstract

The invention discloses an anti-attack detection method based on a network node topological structure. The method comprises the following steps: S1, importing a network and selecting a node as an attack object; s2, calculating five network topology properties: clustering coefficient, betweenness centrality, approximate centrality, feature vector centrality and neighbor node average value; s4, constructing a feature vector space; s5, attacking the network by using an anti-attack method; s6, extracting five network topology properties from the attacked network and constructing a vector space; and S7, adopting a classifier model random forest in machine learning, and verifying the feature vectors extracted in the S4 and the S6 by adopting a reservation method to obtain classification precision. The invention further provides an anti-attack detection system based on the network node topological structure. According to the method, whether the nodes are attacked by a certain countermeasure attack method or not is detected through the topological properties of the multiple nodes in the network, the complexity of the detection algorithm is reduced, the method is universally suitable for various attack methods, and high detection precision is obtained.

Description

technical field [0001] The invention relates to a network node topology and a method and system for detecting an adversarial attack in the network, in particular to a method and system for detecting an adversarial attack based on a network node topology. Background technique [0002] In recent years, deep neural networks have achieved excellent results in many fields, such as image recognition, natural language processing, etc. Today, deep learning has been applied to many scenarios in our lives, some of which, such as autonomous driving, fraud detection, etc., require deep learning models to have high security and robustness. As a result, many researches on the vulnerability and vulnerability of deep neural networks have also become a hot spot. For example, for image recognition tasks, Goodfellow et al. proposed a gradient-based adversarial sample generation method to make deep neural networks misclassify. These attack methods for deep neural networks show that deep neural...

AI Technical Summary

Problems solved by technology

[0006] In the current research, the existing detection methods are often a relatively complex detection algorithm, which is not very practical in practical applications, and some detection methods are only for some attack methods, and the universality is not good.

Images