Cyber Physical Systems are currently employed in several applications such as healthcare, transport, energy, and industrial systems. However, their communication capability exposes them to several network-level threats. Therefore, ensuring the security of Cyber Physical Systems has become an urgent and vital need. To address this issue, in this work we present a deep learning-based anomaly detection system which exploits a 2D representation of the network traffic. More specifically, we propose to employ an integral transformation of the 2D repre-sentation and a Variational Autoencoder to model the nominal system behavior and identify anomalies under the hypothesis that anomalous samples can not be accurately reconstructed by the model trained on normal data. The achieved results show the effectiveness of the proposed method and pave the way for further research in this direction.
Casarin, S., Baldoni, S., Carli, M., Zanuttigh, P., Battisti, F. (2022). Unsupervised Network Anomaly Detection by Learning on 2D Data Representations. In Proceedings - 2022 9th Swiss Conference on Data Science, SDS 2022 (pp.53-58). 345 E 47TH ST, NEW YORK, NY 10017 USA : Institute of Electrical and Electronics Engineers Inc. [10.1109/SDS54800.2022.00016].
Unsupervised Network Anomaly Detection by Learning on 2D Data Representations
Baldoni S.;Carli M.;
2022-01-01
Abstract
Cyber Physical Systems are currently employed in several applications such as healthcare, transport, energy, and industrial systems. However, their communication capability exposes them to several network-level threats. Therefore, ensuring the security of Cyber Physical Systems has become an urgent and vital need. To address this issue, in this work we present a deep learning-based anomaly detection system which exploits a 2D representation of the network traffic. More specifically, we propose to employ an integral transformation of the 2D repre-sentation and a Variational Autoencoder to model the nominal system behavior and identify anomalies under the hypothesis that anomalous samples can not be accurately reconstructed by the model trained on normal data. The achieved results show the effectiveness of the proposed method and pave the way for further research in this direction.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.