Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

A new outlier detection method based on anti-sparse representations

: Vural, M.; Jung, P.; Stanczak, S.


Institute of Electrical and Electronics Engineers -IEEE-:
SIU 2017, 25. Sinyal Işleme ve Iletişim Uygulamaları Kurultayı = 25th Signal Processing and Communications Applications Conference : 15-18 May 2017, Maritim Pine Beach Resort Belek, Antalya/Türkiye
Piscataway, NJ: IEEE, 2017
ISBN: 978-1-5090-6494-6
ISBN: 978-1-5090-6495-3
Sinyal Isleme ve Iletisim Uygulamalari Kurultayi (SIU) <25, 2017, Antalya>
Signal Processing and Communications Applications Conference (SIU) <25, 2017, Antalya>
Fraunhofer HHI ()

Detecting outliers plays a significant role in many areas such as statistics, machine learning, data-mining, etc. Depending on the importance of accurate outlier detection, they have been investigated extensively in the mentioned fields with many different type of algorithms and approaches. One of the popular one is distance based outlier detection approaches, mostly based on nearest neighbor search. In this study, anti-sparse representation is used for approximate nearest neighbor search. In order to deal with some mathematical difficulties of obtaining anti-sparse vector, a new smooth objective function is presented for (or maximum)-norm minimization problem. The obtained anti-sparse vector is binarized and used for embedded approximate nearest neighbor search for outlier detection. Performance of the new anti-sparse representation based outlier detection method is investigated on a real network data set by detecting denial-of-service (DoS) attacks.