Official implementation of paper "Unsupervised Anomaly Detection with Distillated Teacher-Student Network Ensemble".
Full text: https://www.mdpi.com/1099-4300/23/2/201/htm
A typical command that performs 10 independent runs would be:
python main.py --data-path <your-data-path> --preprocessing minmax --seed 2020 --num-students 8 --num-trans 32 --num-trial 10 --pretrain --classify-score cp --batch-size 128
To see more options, please type python main.py -h
.
If you find our paper is helpful for your research, please cite this paper:
@Article{e23020201,
AUTHOR = {Xiao, Qinfeng and Wang, Jing and Lin, Youfang and Gongsa, Wenbo and Hu, Ganghui and Li, Menggang and Wang, Fang},
TITLE = {Unsupervised Anomaly Detection with Distillated Teacher-Student Network Ensemble},
JOURNAL = {Entropy},
VOLUME = {23},
YEAR = {2021},
NUMBER = {2},
ARTICLE-NUMBER = {201},
URL = {https://www.mdpi.com/1099-4300/23/2/201},
PubMedID = {33561954},
ISSN = {1099-4300},
DOI = {10.3390/e23020201}
}