Mean-Shifted Contrastive Loss for Anomaly Detection
Proceedings of the AAAI Conference on Artificial Intelligence · 26.06.2023 · 37 (2) · 2155-2162
DOI: 10.1609/aaai.v37i2.25309 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 26.06.2023. Verfasst von Tal Reiss und 1 weiteren Autor:innen. Die Arbeit wurde bisher 84-mal zitiert (Crossref).
Abstract (Originalsprache)
Deep anomaly detection methods learn representations that separate between normal and anomalous images. Although self-supervised representation learning is commonly used, small dataset sizes limit its effectiveness. It was previously shown that utilizing external, generic datasets (e.g. ImageNet classification) can significantly improve anomaly detection performance. One approach is outlier exposure, which fails when the external datasets do not resemble the anomalies. We take the approach of transferring representations pre-trained on external datasets for anomaly detection. Anomaly detection performance can be significantly improved by fine-tuning the pre-trained representations on the normal training images. In this paper, we first demonstrate and analyze that contrastive learning, the most popular self-supervised learning paradigm cannot be naively applied to pre-trained features. The reason is that pre-trained feature initialization causes poor conditioning for standard contrastive objectives, resulting in bad optimization dynamics. Based on our analysis, we provide a modified contrastive objective, the Mean-Shifted Contrastive Loss. Our method is highly effective and achieves a new state-of-the-art anomaly detection performance including 98.6% ROC-AUC on the CIFAR-10 dataset.
Zitieren
APA
Reiss, T., & Hoshen, Y. (2023). Mean-Shifted Contrastive Loss for Anomaly Detection. Proceedings of the AAAI Conference on Artificial Intelligence, 37(2), 2155-2162. https://doi.org/10.1609/aaai.v37i2.25309
Vancouver
Reiss T, Hoshen Y. Mean-Shifted Contrastive Loss for Anomaly Detection. Proceedings of the AAAI Conference on Artificial Intelligence. 2023;37(2):2155-2162. doi:10.1609/aaai.v37i2.25309
BibTeX
@article{Reiss2023,
author = {Reiss, Tal and Hoshen, Yedid},
title = {Mean-Shifted Contrastive Loss for Anomaly Detection},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2023},
volume = {37},
number = {2},
pages = {2155--2162},
doi = {10.1609/aaai.v37i2.25309}
}
EUVIMED