Article: Information security based on featureless attack algorithm Journal: International Journal of Security and Networks (IJSN) 2024 Vol.19 No.4 pp.169 - 177 Abstract: This study proposes a model for generating and recovering adversarial samples to address the issue of machine learning systems being vulnerable to attacks. The model includes a featureless attack algorithm based on generative adversarial networks and an adversarial sample generation model. The performance of the machine learning system in the face of attacks is evaluated using this model. To address the issue of poor defence against data processing-based adversarial samples, a convolutional neural network-based adversarial sample recovery model is then built to improve the detection and response capability of machine learning systems facing adversarial attacks. The results indicated that under the attack of the featureless attack algorithm, the accuracy of each classifier gradually decreases, and finally lower than 0.1. The bypass rate of the featureless attack algorithm was high, and the probability of the classifier recognising the antagonistic samples was up to 4%. The convolutional neural network-based antagonistic samples recovery model had better image denoising effect and defence effect. In summary, the model constructed by the research has a good application effect, which helps to improve the antagonistic defence ability of the machine learning system and guarantee the information security. Inderscience Publishers - linking academia, business and industry through research

Title: Information security based on featureless attack algorithm

Authors: Huiru Zhang; Ruixiao Liu; Huijuan Liu

Addresses: Quality Management and Performance Assessment Office, Weifang Engineering Vocational College, Qingzhou, 262500, China ' Party and Government Office, Weifang Engineering Vocational College, Qingzhou, 262500, China ' Scientific Research Industry Division, Weifang Engineering Vocational College, Qingzhou, 262500, China

Abstract: This study proposes a model for generating and recovering adversarial samples to address the issue of machine learning systems being vulnerable to attacks. The model includes a featureless attack algorithm based on generative adversarial networks and an adversarial sample generation model. The performance of the machine learning system in the face of attacks is evaluated using this model. To address the issue of poor defence against data processing-based adversarial samples, a convolutional neural network-based adversarial sample recovery model is then built to improve the detection and response capability of machine learning systems facing adversarial attacks. The results indicated that under the attack of the featureless attack algorithm, the accuracy of each classifier gradually decreases, and finally lower than 0.1. The bypass rate of the featureless attack algorithm was high, and the probability of the classifier recognising the antagonistic samples was up to 4%. The convolutional neural network-based antagonistic samples recovery model had better image denoising effect and defence effect. In summary, the model constructed by the research has a good application effect, which helps to improve the antagonistic defence ability of the machine learning system and guarantee the information security.

Keywords: information security; generative adversarial networks; adversarial samples; convolutional neural networks; CNNs.

DOI: 10.1504/IJSN.2024.143738

International Journal of Security and Networks, 2024 Vol.19 No.4, pp.169 - 177

Received: 10 Jan 2024
Accepted: 22 May 2024
Published online: 06 Jan 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Information security based on featureless attack algorithm

Keep up-to-date