Semi-Supervised Outlier Detection Using A Generative And Adversary Framework

Jindong Gu, Matthias Schubert, Volker TrespZenodo 2018 · DOI

Abstract

In many outlier detection tasks, only training data<br> belonging to one class, i.e., the positive class, is available. The<br> task is then to predict a new data point as belonging either to<br> the positive class or to the negative class, in which case the<br> data point is considered an outlier. For this task, we propose a<br> novel corrupted Generative Adversarial Network (CorGAN). In the<br> adversarial process of training CorGAN, the Generator generates<br> outlier samples for the negative class, and the Discriminator is trained<br> to distinguish the positive training data from the generated negative<br> data. The proposed framework is evaluated using an image dataset<br> and a real-world network intrusion dataset. Our outlier-detection<br> method achieves state-of-the-art performance on both tasks.
BibTeX
@inproceedings{gu2018semi,
  author = {Jindong Gu and Matthias Schubert and Volker Tresp},
  title = {{Semi-Supervised Outlier Detection Using A Generative And Adversary Framework}},
  booktitle = {Zenodo},
  year = {2018},
  doi = {10.5281/zenodo.1474926},
}

Loading related content...

Figure