Trojan horse attack method based on neural network image classification and related equipment

Abstract

The invention provides a Trojan horse attack method based on neural network image classification and related equipment. The method comprises the following steps: randomly selecting part of training data in a training set; performing poisoning operation on the training data to obtain poisoning data; adversarial disturbance is added into the poisoning data to obtain disturbance data; adding the disturbance data into the training set to obtain a training sample; training a neural network model by using the training sample, and implanting a back door into the neural network model to obtain a poisoning neural network model; and triggering the poisoning neural network model by using the disturbance data to realize Trojan attack. The Trojan attack method based on neural network image classification is provided for a deep neural network Trojan attack and defense technology platform, and is used for actually displaying the attack process of the Trojan and verifying the Trojan defense effect.

Description

technical field [0001] The present application relates to the technical field of machine learning, in particular to a method for attacking Trojan horses based on neural network image classification and related equipment. Background technique [0002] As the most important model in the field of modern artificial intelligence, deep neural network has been widely used in the fields of industry, medical treatment, multimedia, World Wide Web and science, and has shown performance close to or even surpassing human intelligence, and its influence even penetrates into politics and social sciences. and the legal field. [0003] However, since the interpretability of the deep neural network as the foundation of the model is still a difficult theoretical problem, the application based on the deep neural network, especially the image classification task it is good at, is difficult to deploy in the environment with extremely high security requirements. in key systems. Its existing nega...

AI Technical Summary

Problems solved by technology

[0003] However, since the interpretability of the deep neural network as the foundation of the model is still a difficult theoretical problem, the application based on the deep neural network, especially the image classification task it is good at, is difficult to deploy in the environment with extremely high security requirements. key system

Images