An Adversarial Sample Attack for ML-Assisted Network Automation
PAN Xiaoqin
YIN Hui
CAI Yi
DUAN Kangrong
Abstract：This paper proposes a black-box adversarial attack scheme to mislead the machine learning(ML)based classifiers for anomaly detection to output incorrect classification results in network automation.Firstly,an adversarial sample generation algo-rithm is designed to synthesize training data for the substitute classifier.The algorithm generates synthetic data to cover all the anom-aly types based on a set of legitimate telemetry data that only containing the"normal"type,and label the synthesized data with mini-mized queries to the legitimate classifier.Then,the generated adversarial sample is leveraged to attack the ML-based classifier to analyze the influence of the adversarial samples on performance of the model,and extend the results to different types of models.Fi-nally,extensive simulations are conducted with the telemetry data collected from a real-world IP over elastic optical network(IP-over-EON)testbed.The results show the effectiveness of the proposed scheme.
Keywords：network automation(NA)black-box attackmachine learning(ML)anomaly detectionadversarial samples
Publication Date：2023-12-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：7( 2821-2826,2858 )