A black-box adversarial example generation method based on low query image data

Abstract

The invention discloses a method for generating black-box adversarial samples based on low-query image data, belongs to the technical field of adversarial sample generation methods, and solves the problem that existing effective black-box adversarial samples have many iterations. In the application of image recognition, the present invention obtains the current image data; randomly selects any known model, sets the white-box confrontation sample generation parameters based on the known model, and then generates the effective white box of the known model based on the white-box generation method and the current image data. Box confrontation samples; subtract the current image data from the white box confrontation samples, and perform a normalization operation to generate the confrontation vector; send the current image data and the current image data with the confrontation vector into the unknown model respectively, After the output is obtained, use the ADAM algorithm to update the current image data. After the update, if the requirements are met, a black-box adversarial sample is obtained. If the requirements are not met, the updated image data is repeatedly executed. The present invention is used to generate black-box adversarial samples.

Description

technical field [0001] A black-box adversarial sample generation method based on low query image data is used for generating black-box adversarial samples, and belongs to the technical field of adversarial sample generation methods. Background technique [0002] In recent years, artificial intelligence technology represented by deep learning has developed rapidly and has been widely used in various fields. Unmanned and intelligent is bound to be an important trend in the future. At the same time, deep learning is also facing serious reliability problems. Recent studies have found that artificial intelligence algorithms represented by deep learning may be attacked maliciously, whether in the experimental environment or in the real physical world. Today, with the increasing application of artificial intelligence, the threat of security vulnerabilities is becoming increasingly severe. Therefore, the security of artificial intelligence algorithms is a key issue to be solved in...

AI Technical Summary

Problems solved by technology

[0005] In view of the above-mentioned research problems, the purpose of the present invention is to provide a method for generating black-box adversarial samples based on low-inquiry image data, which solves the problem of many iterations and low success rate of effective black-box adversarial samples in the prior art

Images