A Nonnegativity-Constraint Sparse Stacked Denoising Autoencoder for Anomaly Detection in Electric Power Communication Network

As the scale of the electric power communication network is getting larger and larger, network security issues are becoming more and more complex. as an effective protection method, network traffic anomaly detection can provide important technical support for network situation awareness. In some cas...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:IEEE International Symposium on Broadband Multimedia Systems and Broadcasting s. 1 - 6
Hlavní autori: Tao, Zhuo, Yan, Yong, Yang, Yang, Wang, Yuanyuan, Luo, Jiang
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 27.10.2020
Predmet:
Anomaly detection
Communication networks
few labeled data
Multimedia communication
Noise reduction
nonnegativity-constraint sparse stacked denoising autoencoder
parameter transfer learning
Power systems
Security
Telecommunication traffic
traffic anomaly detection
ISSN:2155-5052
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:As the scale of the electric power communication network is getting larger and larger, network security issues are becoming more and more complex. as an effective protection method, network traffic anomaly detection can provide important technical support for network situation awareness. In some cases, plenty of labeled data are unavailable, which may lead to low detection accuracy. To deal with this problem, a nonnegativity-constraint sparse stacked denoising autoencoder(NSSDAE) is proposed. Meanwhile, we propose a method called dynamic parameter freezing(DPF) for parameter transfer, which allows to find the best performance that may exist between the freezing and fine-tuning within a layer by adjusting the variable alpha. And discriminative joint probability maximum mean discrepancy is introduced for distribution adaptation. The results for NSL-KDD show that the proposed NSSDAE-TL algorithm is effective.
ISSN:2155-5052
DOI:10.1109/BMSB49480.2020.9379595