RSRAE

Robust Subspace Recovery Layer for Unsupervised Anomaly Detection

Getting Started

This repo implements the main experiments of our ICLR 2020 paper: https://openreview.net/forum?id=rylb3eBtwr

Prerequisites

• python
• tensorflow
• keras
• scikit-learn

Run the following to get the results in the paper.

python experiments.py -t caltech101 -l l21 -q 1 -r 5 -m 10 -z 1
python experiments.py -t fashion -l l21 -q 1 -r 5 -m 10 -z 1
python experiments.py -t 20news -l l21 -q 1 -r 5 -m 10 -z 1
python experiments.py -t reuters -l l21 -q 1 -r 5 -m 10 -z 1

Links for preprocessed data:

Caltech101: https://drive.google.com/open?id=1qSGEy3EMwD7XcndMjJvb-TexAaDwWBy8

20 Newsgroup: https://drive.google.com/open?id=1Cc53BQF39XT0VWJhDTaRBUopQ6Vg75q_

Reuters: https://drive.google.com/open?id=11Jg7UewpVRwp40V5yTuKeeARS0K2XMzF

Tiny Imagenet: https://drive.google.com/open?id=1jfceyTY4SSBxxHE2mohy2rwSjmCZTgNE

Citation

The following is the bibtex for citation.

@inproceedings{lai2020robust,
title={Robust Subspace Recovery Layer for Unsupervised Anomaly Detection},
author={Chieh-Hsin Lai and Dongmian Zou and Gilad Lerman},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=rylb3eBtwr},
}