A method and system for generating adversarial samples of network traffic based on time series

Through the time series-based network traffic adversarial sample generation method, the problem that adversarial sample generation in the prior art is difficult to maintain functionality and applicable to multiple types of traffic is solved, and an efficient adversarial attack effect is achieved.

CN118659907BActive Publication Date: 2025-03-21CHENGDU UNIV OF INFORMATION TECH
1 Cites 0 Cited by

Patent Information

Application Number
CN202410844264.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-27
Publication Date
2025-03-21
Estimated Expiration
2044-06-27

AI Technical Summary

Technical Problem

The existing adversarial sample generation methods are difficult to maintain the functionality of network traffic when generating adversarial samples, and cannot be effectively applied to multiple types of network traffic.

Method used

The network traffic adversarial sample generation method based on time series is adopted to preprocess network traffic data, extract time series features, and generate adversarial perturbations using the 1D-CNN model and the FGSM attack algorithm to ensure the functionality and multi-type adaptability of the adversarial sample.

Benefits of technology

It realizes that effective adversarial samples are generated without understanding the detailed parameters of the classifier model, which significantly reduces the detection accuracy of the model and improves the effect of adversarial attacks.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present invention belongs to the field of artificial intelligence security, and discloses a method for generating adversarial samples of network traffic based on time series. The present invention first uses the time series feature data of network traffic to train a one-dimensional CNN classification network, and then uses the FGSM attack algorithm on CNN to obtain adversarial samples, and injects the adversarial samples as universal adversarial perturbations into the original sample at the location where the network traffic packet transmission direction is exchanged, and then returns the perturbed samples to CNN and uses the FGSM attack algorithm to iteratively generate adversarial samples, and finally injects the generated universal adversarial perturbations into other time series data sets and performs adversarial sample attacks on stacked denoising autoencoders. The present invention can effectively generate universal adversarial perturbations for network traffic time series features, and the universal adversarial perturbations can effectively reduce the classification performance of the model for five types of network traffic: botnets, brute force cracking, port scanning, Web attacks, and DDoS attacks.
Need to check novelty before this filing date? Find Prior Art