Depth map convolution model defense method based on generative adversarial network

Abstract

The invention discloses a depth map convolution model defense method based on a generative adversarial network. The method comprises the steps: (1) building a multi-strategy generator and a similaritydiscriminator based on the generative adversarial network, wherein the multi-strategy generator comprises a graph feature extractor for extracting low-dimensional embedded features of an original graph sample and a graph reconstructor for reconstructing an adversarial graph sample based on the low-dimensional embedded features and a dimension expansion matrix, and the similarity judger is used for judging the real probability of the input image sample; (2) performing adversarial iterative training on the multi-strategy generator and the similarity discriminator to generate an adversarial graph sample which has an attack disturbance effect and is highly similar to the original graph sample; (3) constructing a defense classifier based on the graph convolution network, performing pre-training and defense training on the defense classification model by utilizing the original graph sample and the adversarial graph sample, and taking the defense classifier after training as a final defensemodel; (4) inputting the graph sample into a final defense model to realize identification and classification tasks with defense and countermeasure attacks.

Description

technical field [0001] The invention belongs to the technical field of network security, and in particular relates to a defense method of a deep graph convolution model based on a generative confrontation network. Background technique [0002] Modern life is surrounded by various network data, which are used to represent data in many fields, such as social network, communication network, biological network, transaction network, etc. In order to express complex and diverse network data in an intuitive data form, researchers mostly describe network data in the form of graph data. Graph embedding methods map the node and edge information in a graph to a low-dimensional Euclidean space, thereby enabling graph analysis tasks in the real world. Deep graph convolutional models, one of the most successful graph embedding methods, have shown encouraging results in various applications, such as node classification, graph classification, link prediction, and community detection. Low-...

AI Technical Summary

Problems solved by technology

However, in most cases, this randomness is difficult to guarantee the effectiveness of the defense. How to more targetedly defend against the adversary's attack on the network, improve the robustness of the depth map convolution model, and improve the performance of downstream tasks has important practical significance

Images