Generation of Industrial Protocol-Traffic via Enhanced Wasserstein GAN
Authors: Iker Pastor
Date: 08.10.2024
Abstract
In recent years, due to the increase in the number of smart devices connected to the internet, IoT system security has quickly become an issue of societal concern. Because of the large volume of data generated by the IoT platform, traditional protection mechanisms may be ineffective. Thus, new machine learning (ML) solutions have been introduced. However, in the cybersecurity context, both the lack of available data and the privacy issues are recurrent problems when tackling network anomaly and attack detection tasks. Recent works have shown that synthetic data obtained from generative models can help mitigate this issue for network anomaly detectors. Based on this, the following work applies the power of generative techniques in an industrial context to achieve high-quality synthetic data. The obtained results show that the synthetic and real data distributions are similar enough to be used in model training or injected as background noise in real networks.
BIB_text
title = {Generation of Industrial Protocol-Traffic via Enhanced Wasserstein GAN},
pages = {88-97},
keywds = {
Generative Adversarial Network; Internet of Things; Network Traffic; Synthetic Data Generation; Unsupervised Learning
}
abstract = {
In recent years, due to the increase in the number of smart devices connected to the internet, IoT system security has quickly become an issue of societal concern. Because of the large volume of data generated by the IoT platform, traditional protection mechanisms may be ineffective. Thus, new machine learning (ML) solutions have been introduced. However, in the cybersecurity context, both the lack of available data and the privacy issues are recurrent problems when tackling network anomaly and attack detection tasks. Recent works have shown that synthetic data obtained from generative models can help mitigate this issue for network anomaly detectors. Based on this, the following work applies the power of generative techniques in an industrial context to achieve high-quality synthetic data. The obtained results show that the synthetic and real data distributions are similar enough to be used in model training or injected as background noise in real networks.
}
isbn = {978-303175015-1},
date = {2024-10-08},
}
